Two hydrological stations were established in May 2015: one at Carnivore Creek (Schiefer et al., 2019a; Fig. 6a) and one at Chamberlin Creek (Schiefer et al., 2019b; Fig. 6b). At each station, an In-Situ TROLL 9500 (TROLL) was deployed during the open-channel season to measure pressure, temperature, conductivity, and turbidity, and a Hobo U20 water level logger was used for backup measurements of pressure and temperature. The instruments were attached to a cage constructed from aluminum tubing and chicken wire. A stage gauge was installed at each of the hydrological stations, and water level and reach conditions were captured in hourly photographs by field cameras. Cages were installed in thalwegs of stable reaches and loaded with boulders to resist movement (Fig. 6b). The field procedure was repeated for the 2016, 2017, and 2018 field seasons (May–August), though no instruments were successfully deployed in Chamberlin Creek in 2018, and only Hobo data (not TROLL) were recovered from Carnivore Creek in 2018.

In Carnivore Creek, the cage for the hydrological station was secured in the same location in all years in the thalweg of the main channel, near the outlet to Lake Peters, but upstream of where the channel anabranches on the alluvial plain (Fig. 6a). In contrast, in Chamberlin Creek, the location of the hydrological station was modified from year to year due to the difficulty of securing the TROLL and Hobo loggers to withstand the highest flood events of each field season. For 2015, the Hobo data provide a more complete record of water pressure measurements at Chamberlin Creek, and were therefore primarily used for the continuous time series. However, on 4 June the Hobo sensor failed and was re-established in a slightly different position on 5 June. The small shift in instrument position was corrected by comparing water pressure from the Hobo and TROLL and adjusting the early season Hobo water pressure from 5 June through the end of the season. Following these issues at Chamberlin Creek in 2015, the instruments were secured to channel-side bedrock in an upstream pool for the 2016 field season, but were ultimately carried downstream in the largest flood event of 2016 on 8 July. For the 2017 field season, the instruments were secured to a large boulder near the original 2015 hydrological station.

In addition to the continuous measurements from the TROLL and Hobo loggers, discrete discharge (Q) and suspended sediment concentration (SSC) measurements were taken using a hand-held Hach FH950 flow meter and a US DH-48 suspended sediment sampler, respectively, during the 2015, 2016, and 2017 field seasons. Discrete field sampling of discharge was most frequent and spanned the greatest range in 2015. In Carnivore Creek, discrete measurements were taken adjacent to the hydrological station using a cross section with a seasonally deployed tag line to measure consistent 1 m widths between observation points across the creek in the same location each year. In Chamberlin Creek, discrete measurements were primarily taken at the site of the 2015 hydrological station, though coordinates varied depending on the location of the instruments (described above).

Barometric pressure measurements from the 850 m a.s.l. meteorological station were used to convert instrument heights above the riverbed to stage. Discharge–stage regressions were used to construct continuous hourly discharge for the majority of the 2015 open-channel season in Carnivore and Chamberlin creeks (Fig. 7a–b). In Carnivore Creek, photographs of water level were imported to image viewing-and-analysis software to estimate 2.6 d of peak stage during a flood on 3 August 2015 (Fig. 7c) and its error margin. The regression used nine photographs selected to capture maximal water-level variability, and stage was established using a point-to-point method to capture temporal variability. In Chamberlin Creek, linear relations between stage from the Hobo and TROLL instruments permitted the compilation of a seamless discharge record for 2015 (Fig. 7d).

For the 2016 open-channel season, alternative methods were applied due to the lack of velocity data required to compute discharge, assuming consistent discharge–stage relations between seasons (Thurston, 2017). To calculate continuous discharge for Carnivore Creek, stage was first regressed against cross-sectional area for both open-channel seasons separately (Fig. 7e). Stage for 2016 was then adjusted by the mean difference between 2015 and 2016 stage for the same data range for both years (Fig. 7e), to be consistent with 2015 stage by accounting for the shift in instrument positioning between years. The 2015 stage–discharge relation was then used to establish continuous discharge from stage measured in 2016 for that field season (Fig. 7a). The 2015 and 2016 stage data are remarkably consistent between seasons (Fig. 7e), and cross-sectional tag line and instruments were secured in the same locations in 2015 and 2016. However, there is potential for slightly different instrument positions between the two seasons, as well as for bed scour during the August 2015 flood. Therefore, while these data provide a reasonable estimate of 2016 discharge at Carnivore Creek, they do not conform to the standard procedure of obtaining a new rating curve for each season and are therefore considered approximate.

The Chamberlin Creek data for 2016 are also approximate due to both the lack of a new rating curve for this season and the unconventional methods used to calculate continuous discharge. The method used for Carnivore Creek could not be applied to Chamberlin Creek because the TROLL and Hobo instruments were secured in different locations between 2015 and 2016. Instead, 2015 discharge was regressed against cross-sectional area (Fig. 7f), and this relation was then used to construct discrete 2016 extrapolated discharges sampled from 2016 cross-sectional areas. These discharge–stage relations were used to calculate continuous discharge for 2016 until a flood event that began on 20 June (Fig. 7g) washed the instruments downstream. Several methods were used to construct stage and discharge following this flood (Fig. 7h–j). These methods included photographs of river level from time-lapse cameras (46 d of discharge values); a regression of Chamberlin Creek and Carnivore Creek discharges using all available data (3.6 d); and a three-point stage–discharge rating curve sampled downstream of the rating curve applied prior to the flood (8.8 d). When none of the aforementioned alternatives were possible, gaps were interpolated, amounting to a total of five gaps that range from 2 to 19 h. The majority of the 2016 stage record (20 June–5 August 2016) was estimated for Chamberlin Creek from photographs of water levels (Fig. 7h; Thurston, 2017). Water levels were measured using two reference points on a total of 13 photographs, then regressed separately against discharge. The first reference point was suitable for most discharges but was not accurate after 16 July 2016 when water levels dropped low enough to expose a small bar. The second reference point was suitable for low discharges but overtopped during high flows. Error was conservatively 10–20 image pixels, where 10 pixels equate to an approximately 0.14 m3 s-1 error margin in either direction from the average discharge of 0.72 m3 s-1. For the peak of the flood event on 8 July at 02:00 GMT-9, error is estimated to be 100 pixels in either direction because water level and control points were obscured (equating to a maximum error range of 7.31 to 49 m3 s-1 for the peak discharge of 21 m3 s-1). The Chamberlin Creek hydrological station was ultimately re-established in a new location, and a new rating curve was used to reconstruct discharge from 8 August 2016 at 18:00 until 17 August 2016 at 13:00 (Fig. 7i). Where no discharge–stage rating curves or photographs were available, the relation between Carnivore Creek and Chamberlin Creek discharges was used (Fig. 7j). The photographic method was favored whenever photographs were available to keep the sub-catchment discharge records independent. Considerable scatter between Carnivore and Chamberlin Creek discharge records is attributed to hydrological differences between the sub-catchments (Fig. 7j).

Stage–discharge rating curves developed from 2015 and 2016 data were used to convert continuous stage records to discharge for Carnivore Creek for 2017 and 2018, and for Chamberlin Creek for 2017. Adjustments were made to account for changes in instrumentation position relative to the channel bed. The 2017 and 2018 data for Carnivore Creek and the 2017 data for Chamberlin Creek are therefore also considered approximate. While the limitations of the discharge data are recognized, they are still valuable indicative records of hydrological conditions in this remote Arctic setting, where few such datasets exist. Summarized discharge data alongside meteorological data at the 850 m a.s.l. station are shown in Fig. 8.

To estimate the full hydrologic season discharge volumes (Table 1), the continuous Carnivore and Chamberlin Creek time series required extrapolation to include the relatively low late-season flows following removal of instruments. Mean daily discharge for this period was estimated based on the relation between Carnivore and Chamberlin creeks and the nearby Hula Hula River, which explains almost half the variability in our study creeks in the late season.

Hourly water-level measurements were taken using Hobo U20L water level loggers installed on anchored moorings near the water's surface at two locations in Lake Peters: one in its central basin and one near its outflow (Fortin et al., 2019a). Pressure data were corrected to water depth using barometric pressure measurements from the 850 m a.s.l. meteorological station.

In 2015, a series of TROLL casts were taken to measure temperature, turbidity, and conductivity profiles at stations along a north–south transect throughout Lake Peters (Fortin et al., 2019b). These casts were performed at a varying number of stations on a total of 33 d over the course of the field season.

In 2015, 2016, and 2017, lake temperature and luminous flux data were collected at various water depths on anchored moorings throughout the lake (Fig. 2) (Kaufman et al., 2019c). The moorings were equipped with Hobo pendants and Hobo Water Temp Pro loggers at different water depths to monitor the thermal structure of the lake. Moorings located in the central basin of the lake were deployed from May 2015 until August 2016 (Station 7) and August 2015 until August 2017 (Station 8). A mooring in the distal basin (Station 11) was deployed from May 2015 to August 2015, and a mooring in the proximal basin (Station 3) was deployed from May 2015 until August 2015.

Throughout 2015–2017, sediment traps were deployed in Lake Peters to measure the rate of deposition and to collect suspended sediments in the lakes (Kaufman et al., 2019d). In 2015, three pairs of static (i.e., non-automated) traps with different aspect ratios (different collection tube heights but same collection tube diameters) (Fig. 9a) were deployed from May to August at a central location of Lake Peters to determine the effect of these proportions on sediment collection efficiency. From May to August 2016 and again from August 2016 through August 2017, one static trap was deployed in Lake Peters. This trap consisted of 50 mL centrifuge tubes attached to the mouths of inverted 2 L plastic bottles (beverage containers) with the bottoms removed (to funnel sediments into the tubes) (Fig. 9b). Each trap comprised two or three replicate tubes and bottles deployed at different depths along an anchored mooring. In addition, an incrementing sediment trap described by Muzzi and Eadie (2002) was deployed from May 2016 through August 2017, collecting sediments in 23 bottles rotating over daily to weekly increments (Fig. 9c).

For all sediment samples, dry sediment mass, daily flux, and annual flux were calculated. For several static trap samples, grain-size data (mean, standard deviation, d50, d90 and percentages of sand, silt, and clay) were determined using a Coulter LS-320. Samples were pretreated with 30 % H2O2 and heated overnight at 50 ∘C to remove organic material, and then treated with 10 % Na2CO3 for 5 h to remove biogenic silica. Pretreated samples were deflocculated by adding sodium hexametaphosphate and shaking for 1 h before performing grain-size analysis. Organic-matter content for some samples was determined using loss-on-ignition analysis.

Bathymetric data were collected using a Lowrance Elite-5 HDI Combo sonar depth recorder along several transects throughout Lake Peters in August 2015 (Schiefer et al., 2019c). These data were used to create a bathymetric map with ArcGIS 9.1.

Airborne photogrammetric data were acquired two times in 2015 (23 April and 5 July), three times in 2016 (19 April, 11 July, and 5 August), and three times in 2017 (28 April, 27 June, and 3 September) (Nolan, 2019). These data were used to create digital elevation models (DEMs), with a horizontal resolution of 50–80 cm provided in NAD83(2011) NAVD88 Geoid 12B. When conditions permitted (23 April 2015; 5 July 2015; 19 April 2016; 28 April 2017), imagery of the entire Carnivore Creek watershed was acquired. On all other dates, only the eastern side of the valley, where the catchment glaciers are located, was captured. These data were acquired using a Nikon DSLR attached to a survey grade Global Navigation Satellite System (GNSS) on board a manned aircraft, providing photo-center position to 10 cm or better and thus eliminating the need for ground control. Accuracy within this steep mountain environment was found to exceed 20 cm horizontally and vertically (Nolan and DesLauriers, 2016), without the use of ground control. Data acquisition methods are described in detail in Nolan et al. (2015). These DEMs were used to create difference DEMs that can be used to measure snowpack thickness throughout the entire watershed as well as glacier surface elevation change. The accuracy of these DEM-derived snow thickness measurements has previously been found to be comparable to those acquired using hand-operated probes (Nolan et al., 2015). However, larger errors are expected in rugged mountain topography, as the 50 cm DEMs cannot resolve many small crenulations, likely causing some spatial biasing.