All meteorological stations were equipped with a tipping bucket rain gauge (TB4, Hydrological Services America, Lake Worth, USA, and TR-4, Texas Electronics at East Buxton) and an air temperature/relative humidity sensor (H2SC3, Campbell Scientific, Edmonton, Canada) with a solar radiation shield. Continuous measurements of snow depth were taken at three high-elevation stations (449–740 m a.s.l.; SR50A, sonic distance sensor, Campbell Scientific, Edmonton, Canada). Due to access limitations (remote environment and adverse winter weather conditions), snow density observations could not be taken. A custom-made total precipitation gauge (400 mm diameter PVC pipe, 2000 mm tall, with KPSI 700 pressure transducer) was installed at East Buxton and the Reference Station, which has an Alter shield (260-953, Campbell Scientific) to correct for wind-induced undercatch. Wind speed and wind direction were measured at stations with topographic exposure (8 of 14 stations; 05106-10, marine wind monitor, Campbell Scientific, Edmonton, Canada). Incoming solar radiation was measured at the East Buxton station (SP110, Apogee Instruments).

Air temperature, relative humidity, and rain were measured at 2 m a.g.l. (above ground level) for all stations, except East Buxton, where seasonal snowpacks necessitated the rain gauge orifice and air temperature/relative humidity sensor to be installed at 4.5 and 4 m a.g.l., respectively. Total precipitation was measured at 4 m a.g.l. at both East Buxton and Reference Station. Wind speed and direction were measured at 10 m a.g.l. at Reference Station, 8 m a.g.l. at East Buxton, and at 5 m a.g.l. at all other stations. Tipping bucket rain gauges (TBRGs) were field calibrated semi-annually using a field calibration device (FCD-653, Hydrological Services America, Lake Worth, USA) with a 200 mm h-1 nozzle rate. For the field test, the FCD was emptied into the TBRG twice, and the total number of tips were recorded. The current field specification required the TBRG to be within ±6 tips of 202, where 202 tips is the expected number of tips delivered by two applications of the FCD. If the results of the field test exceeded this threshold, then a correction factor was applied to the TBRG to adjust the volume per tip ratio. The air temperature/relative humidity sensors were replaced every 2 years. The sensors were field checked twice per year in between replacements, using a benchmark sensor (H2SC3, Campbell Scientific, Edmonton, Canada). In addition, a second temperature sensor (109, Campbell Scientific) was installed at each site as a check and backup in case of main sensor malfunction. The measurements of the main sensors were corrected if the difference between the benchmark and the main sensor measurements exceeded the sum of their accuracies (±0.2∘ and ±1.6 %).

Stream water level (stage) was continuously measured at each hydrometric station (starting in fall 2013 for watershed 708 and fall 2014 for all other watersheds), and periodic discharge measurements were taken along the range of potential water levels to develop stage discharge rating curves.

Station locations were selected based on channel stability and the potential to measure discharge. Pressure transducers (OTT PLS-L), anchored to bedrock or boulders with cables armoured with steel, were used to measure the stage (0–4 m range, SDI-12). The mean, maximum, minimum, and standard deviation of 5 s sampling intervals were recorded every 5 min. Stand-alone water level recorders (Odyssey capacitance water level recorder, Data Flow Systems Ltd, replaced by HOBO Water Level Data Logger, U20L-04, in September 2018) were installed in proximity to each pressure transducer as a backup in case of sensor or data logger malfunction. The relative location of each pressure transducer was regularly surveyed to a benchmark to help identify potential sensor movements. Stream conditions and streambed morphology were monitored through time lapse cameras pointing at each station's stream cross section and through photos taken during maintenance visits.

Low flows, generally below 0.5 m3 s-1, were measured manually using the velocity area method midsection discharge equation (ISO, 2007) at least once a year during the summer season. Flow velocities, averaged over a 30 s measurement interval, were measured with the Swoffer 2100 propeller-type mechanical current meter (Swoffer Instruments Inc., Seattle, USA) or the SonTek FlowTracker acoustic Doppler velocimeter (ADV, SonTek, San Diego, USA). Manually measuring flows at moderate to high flows was a challenge for multiple reasons because of rapid streamflow responses to rain events (generally under 12 h; Table 4), late fall and winter storm occurrences when field crews were only on site periodically, and safety issues with both accessing the hydrometric stations and taking manual streamflow measurements at high water levels. Therefore, a novel discharge measurement system was designed based on the salt-in-solution method, as described by Moore (2005), in which salt in solution (5 L water to 1 kg salt) is stored in a 1000 L intermediate bulk container (IBC) tote on site, with a pump to pre-mix the salt solution before a measurement. Another pump transfers the solution to a stainless-steel cylinder used to calculate the volume, and from there, the solution is transferred to a dumping mechanism over the stream that holds up to 40 L (Fig. A2). At predefined stages – based on gaps in the stage discharge rating curve – a signal is sent to release the salt solution. The volume of the solution was predetermined (higher volumes for higher stage levels) to induce a clear signal on top of the potential natural river salinity fluctuations but never exceeding the most sensitive toxicity threshold of 400 mg L-1 (Moore, 2004a, b). Upon initiation of the salt solution pump sequence, a second command is sent to a downstream data logger to activate, at a minimum, two electrical conductivity (EC) sensors (Global Water instrumentation, Inc., College Station, USA; T-HRECS Fathom Scientific Ltd), installed at opposing sides of the stream to capture the passing salt wave. Upon completion of the dump sequence, the EC data are transmitted via the telemetry network for data storage and processing. Recharging of the salt solution reservoir was done manually, and a calibration factor (CF), needed to convert EC sensor readings to the relative salinity of the salt solution, was determined prior to refill and after refill, with a goal of two CFs per salt fill. To increase the accuracy, a triplicate reading was taken for each CF, and each salt-in-solution measurement was matched with one set of triplicate CF values based on their sampling date. Finally, discharge was calculated using the salt volume data and relative salinity (Moore, 2005) for each EC sensor and for each CF. This yielded six discharge values per measurement, which were averaged to calculate final discharge.

Measured weather data include air temperature (∘C), relative humidity (%), rain (mm), total precipitation (mm), snow depth (m), wind speed (m s-1), wind direction (∘), and incoming solar radiation (W m-2). Quality assurance procedures involved a combination of visual and automated inspection of the data for inconsistencies, such as sudden large increases in measurements, and outliers. Table A3 details the thresholds used to identify outliers and inconsistencies in the data. All data, except for the snow depth data, were gap-filled using linear regression from nearby station data. For instances when air temperature readings were faulty, relative humidity (RH) was first converted to vapour pressure and was then converted back to RH based on the appropriate temperature from the nearest station showing the greatest coefficient of determination.

Precipitation data required additional quality assurance procedures due to errors introduced by wind. Suspect spikes in rain intensity caused by wind-induced tips were identified through visual inspection initially but, starting from February 2016, were then done through an automated method. When three or more tips occurred within a 5 s scan interval, the data were flagged, assuming that three or more tips within 5 s would yield an extremely unlikely rainfall rate for the area (> 200 mm h-1). All suspect tips were removed and gap-filled. In addition, hourly rain data from WSN626, WSN693_703, WSN703_708, WSN819_1015, Hecate, East Buxton, SSN693, and Reference Station and hourly total precipitation data from East Buxton and Reference Station were corrected for undercatch (Yang et al., 1998) using the wind speed adjustment from Legates et al. (2005). This adjustment includes a station height variable, thus accounting for potential effects of the elevated total precipitation gauges at East Buxton and Reference Station and the rain gauge orifice at East Buxton (i.e., 4 m instead of 2 m). Please note that the precipitation data at the 5 min time step were not corrected for undercatch, as wind data at the 5 min level introduced too much noise to the corrected dataset. Further on, spikes in the snow depth data (> 1 m) that did not correspond to increases in total precipitation were removed and gap-filled, but gaps that were caused by sensor failures were generally too large and were therefore not filled.

Stage data were quality controlled and flagged by visual inspection. Any spikes, drops, or changes in stage not associated with a rain event were further inspected and, where necessary, manually flagged and corrected. As an example, time lapse photos were used to confirm debris blockages to explain and correct unexpected small fluctuations in stage. Data gaps of < 30 min were filled using linear interpolation (computer automated), while gaps of > 30 min were filled using a relation (linear regression) between data from the backup, standalone water level sensors and pressure transducers. In the one exceptional case where both water level sensors were malfunctioning at watershed 819 (1 December 2015–17 May 2016), the stage records of watershed 844, the neighbouring watershed sharing many catchment characteristics, were used to estimate the stage by linear regression.

Discharge measurement uncertainties were calculated using the interpolated variance estimator (Cohn et al., 2013) for the velocity area method, which estimates uncertainty from the fluctuations in the flow velocity and depth profile and from calibration errors in the width, depth, and velocity measurements. Uncertainty estimations for the automated salt solution discharge measurements were recorded in both a quantitative and a qualitative manner. First, a quantitative percent error was calculated by error propagation of the uncertainty in the volume of salt solution, the EC sensor resolution, and the salt solution electrical conductivity calibration factor (Korver et al., 2018). Measurements with uncertainties higher than 20 % were excluded from further analysis. Second, the EC data were assessed for noise and spikes caused by turbulence at high flows. Where possible, spikes were removed, and noisy data were smoothed. Measurements that contained EC readings so noisy that they could not be corrected were removed. As an example, peak flows at watershed 703 were not successfully measured due to extremely turbulent flow conditions and, hence, extremely noisy EC data. Last, an assessment of salt mixing conditions at the downstream EC sensor site was performed by comparing calculated discharge from the two EC sensors placed within opposing sides of the river. If the calculated discharge differed by more than 1 %, then mixing conditions were considered poor, and unless the measurement filled an essential gap in the stage-discharge rating curve, it was excluded from further analysis.

Each discharge measurement was paired with a time-averaged measurement of the stage (recorded every second for the duration of the measurement), and rating curves were plotted for each station using locally estimated scatterplot smoothing (LOESS) regression (Fig. 2). The span widths of the LOESS fits were visually assessed and selected. Between 5 and 50 new measurements were added to each station's rating curve every year. Measurements prior to and after high-flow events were analyzed for possible shifts in the rating curves; in the case of a shift, time lapse videos and photos taken during maintenance surveys were investigated to confirm a concurrent change in streambed morphology, and a new curve was applied to the discharge time series from the onset of the high-flow event. The curves were extrapolated to minimum and maximum stage of the 5-year stage time series, estimating a zero stage to equal zero flow and estimating maximum flow by extrapolating a power law equation fitted on the upper section of the rating curves.

Following the methodology proposed by Coxon et al. (2015), 95 % confidence intervals (CIs) were plotted around the rating curves in two steps. First, the absolute error of each stage discharge measurement was calculated by error propagation of the discharge measurement uncertainty (described above) and the stage measurement uncertainty. Stage measurement uncertainty was calculated as 2 times the standard deviation of the stage values recorded during the discharge measurement, so that measurements taken during rapidly falling or rising water levels were assigned higher uncertainties. The accuracy of the stage sensors used (±0.2 cm) was found to be inconsequential to the final calculation of discharge uncertainty, compared to flow turbulence and rapidly changing water levels. Second, CIs were derived from 500 curve fitting results of LOESS regressions on 500 randomized sets of stage discharge measurements and their maximum and minimum absolute measurement errors. For each millimetre of stage on the rating curve, 500 discharge values and 500 standard deviations were predicted and combined in a Gaussian mixed model to derive minimum and maximum absolute discharge (95 % CI). Any minimum absolute flow extending below zero was set to zero. Calculations were done in R, and the code has been made publicly available (Sect. 7). The output is a stage discharge lookup table with values of discharge, minimum discharge (95 % CI), and maximum discharge (95 % CI) for each millimetre of stage.

Calvert Island received more precipitation (3353 mm yr-1) than Hecate Island (2676 mm yr-1). At low elevations (< 100 m a.s.l.), precipitation increased along a west-to-east gradient by about 120 mm km-1 on Calvert Island (R2=0.54, for stations between Reference Station and East Buxton) and 150 mm km-1 on Hecate Island (R2=0.88, for stations between SSN1015 and WSN844). Precipitation increased with elevation: about 70 mm per 100 m on Hecate Island (R2=0.26) and about 190 mm per 100 m on Calvert Island (R2=0.54). There was almost a 50 % increase in mean total precipitation between Reference Station on western Calvert Island (42 m a.s.l., 2452 mm) and East Buxton in the east (740 m a.s.l., 3603 mm), with 5 % versus 30 % falling as snow at the respective stations. The seasonal snowpack at East Buxton station reached an average maximum depth of 1.4 m, whereas intermittent snow cover records at lower elevation (WSN693_703; 449 m a.s.l.) and at the south-facing slope of Hecate Island (Hecate; 477 m a.s.l.) never exceeded 0.6 and 0.3 m, respectively.

Air temperature decreased with increasing elevation at a rate of approximately 0.35 ∘C per 100 m (-0.36 and -0.32 ∘C per 100 m on Calvert (R2=0.83) and Hecate (R2=0.87) islands, respectively), from 8.7 ∘C at sea level to 6.0 ∘C at 740 m a.s.l. This lapse rate varied seasonally, with small differences in temperature between low and high elevation stations during summer months (-0.1 ∘C per 100 m in August) and large differences in winter (-0.5 ∘C per 100 m in February). All stations recorded an average relative humidity (RH) between 80 % and 87 %, with high elevation stations reaching lower RH levels (e.g., minimum of 16 % daily RH at East Buxton) than low elevation stations (e.g., minimum of 37 % daily RH at SSN708). General wind directions were uniform across Calvert and Hecate islands, except for the areas around Mount Buxton which showed local variability. The highest wind speed was recorded at Hecate station (27.3 m s-1), which also recorded the highest wind speeds on average (4.4 m s-1), and station SSN693 was, despite being adjacent to a lake, the most sheltered (average of 1.1 m s-1). Station-specific annual weather parameters are summarized in Table 2.

Seasonal variations in weather variables were analyzed for station SSN708, the station with the longest record (installed September 2013; 12 m a.s.l.), as a representative for the study region. There were wet (October–April) and comparatively dry (May–September) seasons, with the dry season starting abruptly in May but transitioning gradually into the wet season in September (Fig. 3). Although the average monthly rainfall was distinctly lower in the dry season (128 versus 320 mm in the wet season), rain events still regularly occurred; the longest dry periods never exceeded between 9 and 11 consecutive days, usually in July or August. The wet seasons were marked by large intermonthly and interannual variations (58–524 mm per month), whereas variations were comparatively small during the dry seasons (22–218 mm per month). November was the wettest and July the driest month of the year (average of 362 and 78 mm, respectively). Frequently recurring high wind storms (> 10 m s-1) from a predominantly southeasterly direction characterized the wet season, and low northwesterly and southwesterly winds (< 10 m s-1) prevailed during the dry season (Fig. 4), averaging each year to 2.7 m s-1. Monthly average air temperature followed an annual cycle ranging between 2.2 ∘C in February and 14.5 ∘C in August, with overall elevated midwinter temperatures recorded in January (4.3 ∘C). This seasonal trend was consistent between years, except for the month of February, which varied markedly from year to year (between -0.8 and 7.6 ∘C). Like air temperature, relative humidity followed an annual cycle, peaking in August (90 %) and reaching a low in February (85 %), with relatively more year-to-year variability in winter.

Between water years, average precipitation and air temperature varied, whereas wind speed and relative humidity were consistent (2.7 m s-1 and 82 %–84 %). Water year 2016–2017 was, within the short record period, relatively wet and cool (7.6 ∘C; 3681 mm), whereas 2018–2019 was drier than average (2770 mm; Table 3). Maximum air temperature was recorded in September 2016 (30 ∘C; SSN693) and minimum air temperature in February 2019 (-16.3 ∘C; East Buxton). Snowfall was highest in the winter of water year 2017–2018 and lowest in 2015–2016, with maximum snow depths measuring 1.8 and 0.2 m, respectively, at East Buxton. Maximum snow accumulation was usually reached in March.