The Arctic is rapidly changing. Outside the Arctic, large-sample catchment
databases have transformed catchment science from focusing on local case
studies to more systematic studies of watershed functioning. Here we present
an integrated pan-ARctic CAtchments summary DatabasE (ARCADE) of
Earth's rapidly changing climate is particularly evident in the Arctic.
Decreasing sea ice extent has amplified Arctic warming, which has led to an
increase in mean land-surface air temperature of 3.1
In the Arctic, marine and terrestrial systems are tightly coupled. More than 10 % of global river discharge flows into the Arctic Ocean (AO), which only contains about 1 % of the global ocean volume (Aagaard and Carmack, 1989; McClelland et al., 2012). In addition, river discharge transports sediment, (organic) carbon, nutrients, and contaminants (Terhaar et al., 2021) into the AO. Arctic rivers integrate over local to regional scales and are therefore useful for studying the impacts of environmental and climatic change at various scales (Holmes et al., 2012).
Permanently frozen soils (permafrost) that are rich in organic carbon (OC) (Hugelius et al., 2014; Mishra et al., 2021) underlie about 60 %–80 % of the AO watershed (Zhang et al., 2000, 2005; Obu et al., 2019). Permafrost conditions have long stabilized the subsurface, but ground temperatures are now warming across the Northern Hemisphere (Biskaborn et al., 2019). Permafrost degradation occurs slowly through deepening of the active layer (the layer that thaws during summer and refreezes during winter) (Ran et al., 2022) or more quickly through abrupt thaw of permafrost with high ground ice contents. Both types of thaw expose soil OC to degradation, which transforms it into greenhouse gases. Thus, the thawing of permafrost can accelerate global warming but also impacts hydrological, biogeochemical, and ecological processes in Arctic ecosystems, with complex consequences for lateral transport of terrestrial material to downstream freshwater and marine systems (Vonk and Gustafsson, 2013).
Investigations of Arctic change (e.g., Schuur et al., 2015; Walvoord and
Kurylyk, 2016; Liljedahl et al., 2016; Lafrenière and Lamoureux, 2019;
Bruhwiler et al., 2021) critically rely on data. The “Arctic Great Rivers
Observatory” initiative, which has run since 2003, is a unique dataset
covering the six largest Arctic rivers (McClelland et al., 2008,
Small and medium-sized watersheds drain roughly a third of the circumpolar landmass (Holmes et al., 2012). In contrast to the watersheds of the six largest Arctic rivers (Ob', Yenisey, Lena, Kolyma, Mackenzie, Yukon), the smaller watersheds are almost exclusively underlain by continuous permafrost (Holmes et al., 2012) and are often directly located at the coast. This makes these small watersheds fundamentally different from “The Big Six” because large rivers drain to a few coastal locations (Mann et al., 2022), while the cumulative inputs of small watersheds are spread over a much larger coastal area. In addition, given their size and proximity to the AO, the changes in these watersheds could be more rapidly transferred and substantial to the Arctic coastal ecosystem.
Outside of the Arctic, the emergence of large-sample catchment databases (e.g., Hartmann et al., 2014; Newman et al., 2015; Alvarez-Garreton et al., 2018), which combine data from many watersheds, have transformed the field from placing emphasis on local case studies towards more systematic insights into drivers of watershed functioning. For example, large-sample watershed studies allow one to reveal regional differences (and similarities) in hydrological response, make space-for-time transformations, and systematically test hypotheses. This has proven critical in, for example, understanding the impacts of climate change (e.g., Berghuijs et al., 2014) and testing modeling implications (e.g., Knoben et al., 2020). Such developments have not yet been possible in the Arctic, as large-sample databases of smaller watersheds are not yet available.
Here, we present an integrated pan-ARctic CAtchments summary DatabasE
(ARCADE) of
The ARCADE database encompasses all major and minor drainage basins that are
considered part of the pan-Arctic watershed, with their outlets draining into the Arctic Ocean
and surrounding seas. More specifically, this includes all watersheds with a
Strahler order of 5 (i.e. at least five hierarchical branching orders) or
larger that drain into the Arctic Ocean, as well as basins that drain into the
Bering Sea and north of the Yukon River outlet, with inclusion of the Yukon
River. This follows the pan-Arctic watershed definition as defined by
McGuire et al. (2009), with an area of 20.4
Circumpolar map of all ARCADE watersheds, 1 km
Terrain parameters such as altitude, slope, aspect, topographic position
index, and slope length and steepness factor (LS-factor) (Renard et al.,
2017) were derived and calculated from Copernicus DEM GLO-90, a high-quality
global 90 m resolution digital elevation model provided by the European
Space Agency (ESA, 2021). The Copernicus DEM was accessed on 10 September 2021. For computational practicality, we chose the 90 m resolution product
rather than the 30 m resolution product. The latter could be used for future
version updates of the ARCADE database. However, we deem the 90 m resolution
sufficiently detailed for our purposes (gaining insights into drainage areas
on a pan-Arctic scale). We constrain the number of catchments in the
database by using Strahler order 5 as the minimum outlet order and 1 km
The DEM was hydrologically conditioned (a.k.a. pit filling) before deriving flow direction, flow accumulation, Strahler order, watershed delineations, and topographic wetness index. This was done using the “r.hydrodem” module (Lindsay and Creed, 2005) in GRASS GIS (Neteler et al., 2012).
We delineated the watersheds at 90 m resolution for subdivisions of the
pan-Arctic landmass using the hydrologically conditioned DEM. This
subdivision was necessary because processing the DEM in one piece was
computationally too intensive. Delineation was done using SAGA GIS (Conrad
et al., 2015) using the module “Channel Network and Drainage Basins”. A lower threshold of Strahler order 5 was
chosen to constrain watershed generation, i.e., only watersheds of streams
with Strahler order 5 or higher at the outlet were delineated. This
threshold was necessary to limit the number of watersheds in the final
product and to ensure that only watersheds with actual streams were
included. Another consideration was that, as the watershed area approaches
the DEM's source resolution, the relative accuracy decreases. Subdivisions
of the pan-Arctic watersheds were combined into one dataset of all
watersheds that drain into the AO (i.e., upstream areas of outlets at the
AO). A known limitation of DEM-derived watershed delineation is that the
algorithm struggles to find the channels and ridges in flat terrain. Since
we are mostly interested in the drainage area rather than channel location,
errors in channels were tolerated more than errors in catchment boundaries.
Small, flat catchments (area
All variables are described in file S1 in
Climatological data were extracted from the ERA5-Land monthly averaged –
ECMWF climate reanalysis dataset (Muñoz-Sabater et al., 2021) using
Google Earth Engine (“Planetary-scale geospatial analysis for everyone”; Gorelick et al., 2017). This dataset has a spatial resolution of 11 132 m and consists of 50 bands containing climatological variables related
to temperature, precipitation, evaporation, heat fluxes, wind, and
vegetation. Minimum, maximum, mean, standard deviation, and median annual
values of a subset of these variables (a complete overview of all variables
is available in file S1) were calculated for each watershed from 1 January 1990 to
31 December 2019. In the case of pixels falling partly within the geometry of a
watershed, the value is weighted by the fraction of each pixel that falls
within the geometry. Precipitation, evaporation, and runoff totals were
accumulated and averaged over the 30-year period (i.e., the mean annual
total of each of these variables was calculated). For snow statistics, we
calculated the 30-year average maximum snow depth (m), snow cover (%),
snowmelt (m d
We also tested for trends using Sen's slope estimator for the same period.
Sen (1968) calculates the slope as
Basic catchment properties include minimum, maximum, mean, standard deviation, and median of elevation (meters), slope (degrees), and aspect (degrees). Furthermore, we included centroid latitude (degrees), Gravelius index (watershed perimeter divided by the perimeter of a circle that has the same area; unitless), watershed perimeter (kilometers), and watershed area (square kilometers).
SoilGrids is a globally consistent dataset that contains soil properties
(soil organic carbon, SOC, content – dg kg
Watershed land cover fractional coverage was obtained from ESA WorldCover 10m v100 (Zanaga et al., 2021). This classifies the land surface at 10 m resolution into 11 classes: trees, shrubland, grassland, cropland, built-up areas, barren or sparse vegetation, snow and ice, open water, herbaceous wetland, mangroves, and moss and lichen.
Another useful characterization parameter for watersheds is the fractional
coverage of landforms. We chose to use a landform classification scheme
proposed by Theobald et al. (2015). Their classification scheme maps
ecologically relevant landforms (see tables included in
dataset file S1 in
The burned-area fraction for each watershed over the period 2012–2022 was
calculated from MODIS FireCCI5, a monthly global 250 m spatial
resolution burn scar classification product (Padilla Parelada, 2018). We
selected and summarized recent (
Permafrost fraction pixel cover was taken from the permafrost extent by Obu et al. (2019) and converted into watershed area fractional coverage per permafrost coverage type. The used product has a spatial resolution of 1 km and a temporal range from 2000–2016. Continuous permafrost is classified as a pixel area coverage of 90 %–100 %, discontinuous permafrost as 50 %–90 %, sporadic permafrost as 10 %–50 %, and isolated patches of permafrost as 0 %–10 %.
Recently published high-resolution estimates of active-layer thickness (ALT)
(Ran et al., 2022) were summarized for each watershed. The source dataset
has a 1 km resolution for the period of 2000–2016. The authors
generated the data by combining large amounts of field data and
multisource geospatial remote sensing data into a statistical learning
model. It has bias
The distribution of watershed areas in the pan-Arctic watersheds
database and the range of the four groups that are classified based on watershed
area. “BS” stands for “Big Seven”, “MN” for “Middle Nine”, “PAT” for
“Pan-Arctic Thousands”, and PAS for “Pan-Arctic Small watersheds”. Note that
the
Glacial coverage was calculated by combining two datasets: Global Land Ice Measurements from Space (GLIMS), from which we used the latest available snapshot as of 14 September 2021 for the glacial extent (Kargel et al., 2014), and the Greenland ice and ocean mask from the Greenland Mapping Project (GIMP), which contains a 15 m resolution land ice mask for the Greenland ice sheet (Howat et al., 2014). We resampled the combined datasets to a 250 m resolution grid to calculate fractional glacial coverage for each watershed.
A high-resolution water mask, JRC Global Surface Water Mapping Layers, v1.3 (30 m) (Pekel et al., 2016), was used to calculate fractional watershed area coverage. The conditions for the presence of water were determined by the occurrence of water in each cell for at least 50 % of the time between 1984 and 2020.
The summarized statistics of the normalized difference vegetation index (NDVI) and the Sen slope of NDVI were calculated using MOD13A1.006 Terra Vegetation Indices 16-Day Global 500m (Didan, 2015). This dataset is MODIS derived and has a 500 m resolution. We used the annual maximum NDVI of each year from 2000 to 2021.
As an indicator of terrain wetness, we used SAGA wetness index (Böhner and Selige, 2006), a modified topographic wetness index that is based on Moore et al. (1993). The indicator uses topography to differentiate catchments dominated by wetland terrain versus more well-drained terrain.
Slope length and steepness factor (LS-factor) is a factor used in the Universal Soil Loss Equation (USLE) (Renard et al., 2017) that serves as a predictor of soil loss ratio as a function of slope length and steepness. The LS-factor was calculated using the SAGA GIS tool module “LS-factor”, which uses specific catchment area (SCA) as a substitute of slope length (Böhner and Selige, 2006).
As an indicator for changes in wetness (TCW), greenness (TCG), and brightness (TCB) (indicative of bare soil), we included tasseled-cap indices derived from Landsat visible-spectra images, as provided by Nitze et al. (2018). The minima, maxima, and average of these pixel-based slopes were calculated for each watershed.
The database consists of 47 054 watersheds ranging in size from 1
to 3.1
Summary statistics of the pan-Arctic watersheds database focused on permafrost. Note that this summary excludes watersheds that are fully covered by glaciers or ice sheets.
The BS account for 50 % of the total AO watershed area, while watersheds
under 1000 km
Watershed topographic properties summarized by group (classification based on area) and relevant (sub)continent.
Watershed permafrost properties summarized by group (based on area) and relevant (sub)continent.
The hydrological functioning of small catchments in the Arctic remains
uncertain. Therefore, our database provides a set of catchment properties to
help address these uncertainties. Basic topographical catchment metrics such
as area, elevation, catchment slope, mean aspect, LS-factor, and TWI are
available for all recorded catchments. Due to their resolution and extent,
some of the other aggregated datasets have lower coverage. Most notably,
ERA5-Land data has
The ARCADE database is the first published 90 m resolution dataset of watersheds draining into the Arctic Ocean. A few unavoidable errors occurred during the watershed delineation. Errors most commonly arise in flat terrain where flow-routing algorithms struggle to determine the flow direction, which troubles the watershed border definition. To deal with this, we used an internal SAGA function to artificially maintain a minimal channel slope by slightly altering the DEM. This minimal slope function effect is visually detectable in small deltas and floodplains where watershed borders sometimes appear to be less accurate than in steep, well-defined terrain. Additionally, this flow path uncertainty in flat terrain caused some errors in approximating the locations of coastal outlets. Given the high DEM resolution, these errors are generally in the order of meters rather than kilometers. This could be improved in future versions by “burning” outlets and channels, such as those derived from satellite imagery, into the DEM.
Our cut-off value in defining a river catchment (outlet Strahler order 5;
minimum area of 1 km
Siberian coastal watersheds with ice wedge polygon (IWP) terrain
(% watershed coverage)
Correlations of binned data of selected catchment properties from our database. We calculated Spearman's rho on the binned data. Most notably, we observe that small watersheds have experienced the greatest warming while having the highest mean carbon stocks and the highest fraction of IWP terrain. Similarly, the data show that high OC stocks are found where most warming has occurred.
ARCADE provides 103 variables with catchment properties divided over 353 columns (including statistics), showcasing a wide variability and spatial resemblances of catchments in the pan-Arctic drainage basin. Additionally, we provide summaries of the most important properties for the BS, MN, PAT, and PAS, both as a whole and on a regional basis (i.e., North America, Greenland, and Eurasia).
Basic catchment-scale topographical information can be used to categorize
watershed types and to estimate their runoff, sediment transport regimes, and
biogeochemical constituents. As an example, Connoly et al. (2018) found
strong negative correlations between catchment slope, DOM, and NO
Watershed land cover type and properties summarized by group (classification based on area) and relevant (sub)continent.
Watershed climatological properties summarized by group (classification based on area) and relevant (sub)continent.
Since MN, PAT, and PAS are, on average, located in higher latitudes, these
watersheds are colder than the BS (BS:
We provide this database as a basis to explore the vast number of watersheds outside the BS and MN that have previously been lumped into a single “unknown”. As a result, they have been underappreciated in terms of their contribution to the pan-Arctic lateral flux budget and their potential sensitivity to climate change as opposed to their bigger siblings. While continuing the scientific focus on large catchment studies (BS) in the Arctic remains vital, we suggest, in parallel, to strongly increase the focus on pan-Arctic small catchments situated entirely at high latitudes. These catchments are experiencing the greatest climatic warming while also storing large quantities of soil carbon in landscapes that are especially prone to degradation of permafrost (i.e., IWP terrain) and associated hydrological-regime shifts. Using our database, these and many other variables are now quantified and made spatially explicit (Figs. 3, 4).
The ARCADE database is publicly available via DataverseNL:
ARCADE is the first aggregated database of pan-Arctic river catchments that includes small watersheds at a high resolution. The publication of this database is a necessary step toward more integrated monitoring of the pan-Arctic watershed. An important addition in the following version will be discharge data and derived seasonality (and changes therein) from the RADR database (Feng et al., 2021), which recently greatly advanced understanding of discharge in smaller Arctic rivers. Another important future addition will be the delineations of subbasins and data on river biogeochemistry that is available, albeit non-uniformly and largely unaggregated throughout literature. When numerous valuable datasets from various scientific disciplines are merged, it will be possible to better understand the Arctic's changing hydrology and biogeochemistry. This allows the scientific community to form new hypotheses that direct scientific efforts to specific regions and processes that may have remained under the radar.
NJS coordinated the ARCADE database creation and structure; constructed the data-processing pipeline of the database; collected, aggregated, and inserted all data records; and wrote the first version of the paper. HL and JEV were involved in the initial conceptualization of the database. JEV, WRB, AP, HG, TG, and PAP provided insights from their respective fields of expertise to guide the following database conceptualization steps and during various iterations of its creation. All authors contributed to the final draft of this paper through their various fields of expertise. JEV and HL played key roles in providing funding for the project.
The contact author has declared that none of the authors has any competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The authors thank all those who have made this work possible. We thank Justine Ramage for hosting working sessions at Stockholm University. Ingmar Nitze is thanked for providing helpful advice for using Google Earth Engine. We thank Caroline Coch for her role as a sparring partner at the very beginning of this project.
This research has been supported by the Horizon 2020 program (Nunataryuk (grant no. 773421)), and additional financial support was received from ERC (THAWSOME (grant no. 676982).
This paper was edited by Lukas Gudmundsson and reviewed by Lucas Menzel and one anonymous referee.