2. METHODS
Como Creek (Figure 1) is located in the Niwot Ridge Long Term Ecological Research (LTER) site in the state of Colorado. Located just east of the North American continental divide, it has an area of approximately 5 km2, with elevations ranging from 3,000-3,600 m above sea level. This makes it one of the highest instrumented catchments anywhere in the world. It experiences long cold winters, and short cool summers; mean air temperatures for January are -12ºC, and mean temperatures for July are 12ºC. The average annual precipitation is 730 mm, with roughly two thirds as snowfall (snowfall values are reported in snow-water equivalent; SWE). The majority of the watershed is forested, consisting of Engelmann spruce (Picea engelmannii), subalpine fir (Abies lasiocarpa), limber pine (Pinus flexilis), lodgepole pine (Pinus contorta var. latifolia) and quaking aspen (Populus tremuloides). Approximately 10% of the watershed is above treeline, consisting of alpine tundra and scree slopes.
We obtained four years (Oct 2018 - Sep 2021) of high frequency data from the National Ecological Observatory Network (NEON). NEON (https://neonscience.org) is a National Science Foundation-funded network of monitoring sites throughout the United States providing long-term, open-access ecological data (Goodman et al., 2015). Since 2017, NEON has maintained a monitoring reach along Como Creek, instrumented with a standardized suite of automated sensors. Stream stage is recorded using AquaTroll 600 vented pressure transducers (In-situ; Fort Collins CO). Bi-weekly manual Q measurements are used to develop rating curves and estimate continuous Q. Rating curves are created for each water year (defined as October 1 through September 30). Water quality measurements, including specific conductance (SpC), dissolved oxygen (DO), and fluorescent dissolved organic matter (fDOM) are measured at one minute intervals using an EXO2 multiparameter sonde (YSI; Yellow Springs OH). Stream NO3-N is measured using a submersible ultraviolet nitrate analyzer (SeaBird Scientific, Bellevue WA) configured to take a 20 measurement burst at 15 minute intervals. The first 10 bursts of each measurement are discarded to allow the SUNA lamp sufficient time to warm up. Concentrations reported in μM were converted to mg-N L-1 using the molar mass. The sensors remain installed throughout the winter, measuring concentrations in the liquid water under the ice and snow cover. Both the EXO2 and SUNA were equipped with automated wipers to prevent biofouling. They are also manually cleaned bi-weekly, and calibrated monthly. A time-lapse video of the stream using images from the NEON monitoring location is included as Video 1. Additional images, including real-time, are available from the PhenoCam Network (https://phenocam.sr.unh.edu/webcam/sites/NEON.D13.COMO.DP1.20002/).
NEON also collects bi-weekly grab samples of stream water chemistry. Samples are collected and stored on ice until analysis at the EcoCore laboratory at Colorado State University. In addition to major cations and anions, grab samples are analyzed for DOC, total organic carbon (TOC), NO2+NO3-N (of which NO3-N is the overwhelming majority in Como Creek), ammonium (NH4-N), total dissolved nitrogen (TDN), and total nitrogen (TN). Grab samples of these additional species allowed us to contextualize the sensor-based DOC and NO3-N measurements in terms of the total C and N budgets.
We used the neonUtilities R package (Lunch et al., 2021), to download the following publicly available NEON datasets: Continuous discharge (NEON 2021a), Water quality (NEON 2021b), Nitrate in surface water (NEON 2021c), Temperature in surface water (NEON 2021d) and Chemical properties of surface water (NEON 2021e). Quality flagged measurements were excluded from our analysis; this constituted a relatively small fraction of the total data (~5%) and the majority of these were periods in winter when the stream froze to a depth where the sensors became encapsulated in ice and were no longer measuring concentrations in the liquid water beneath. Because they occured when the stream was not flowing, their omission does not substantially impact annual flux estimates. In a few instances, NEON maintenance and calibration records were used to correct for drift or calibration offsets in the data. Datasets published at higher frequencies (e.g. water quality) were averaged to 15 minute intervals to match nitrate in surface water, which had the lowest temporal resolution. A linear regression between bi-weekly manual DOC measurements and corresponding sensor fDOM measurements was used to estimate continuous DOC from the fDOM time-series. Multiplying DOC and NO3-N concentrations by the corresponding Q measurement, we calculated a continuous record of DOC and NO3-N flux. The na.spline function in the zoo R package was used to fill short gaps of less than 6 hours. Two larger concentration gaps (fDOM from 24 May to 4 June 2018 and NO3-N from 27 June to 1 Aug 2019) were filled using linear interpolation. The effects of uncertainty in this concentration approximation on flux values are expected to be relatively small given that Q is expected to be the primary driver. All other major gaps were left unfilled because they occurred during periods of little to no flow and were deemed to exert almost no influence on annual flux budgets.
To quantitatively compare interannual variability, we determined the date of the centroid of the annual melt pulse, the annual water yield (WY; cumulative Q divided by watershed area), and the annual export of DOC and NO3-N. The NEON precipitation gage is located next to the eddy flux tower, high on Niwot ridge, at an elevation much higher than most of the catchment. Instead, we obtained precipitation and snowpack data was from the National Water and Climate Center Snow Telemetry (SNOTEL;https://www.wcc.nrcs.usda.gov/snow/) Niwot station (ID 663), which is located near the center of the Como Creek catchment and likely more representative. It also has a much longer period of record. From these data we calculated the total annual precipitation, maximum depths of annual snowpack, and the date in the spring when the snowpack depth dropped below 10 cm. We then compared these values with the annual melt pulse and solute flux metrics calculated above.
For each solute we determined the coefficients of variation (CV) for discharge and concentration to determine which constituent component of flux exhibited the larger degree of variation, and thus acted as the primary control. We calculated Gini coefficients (G) to quantitatively characterize the temporal inequality in flux (Jawitz and Mitchell et al., 2011). Commonly used to characterize the distribution of wealth, a Gini coefficient of zero represents complete equality (i.e. a constant mass flux rate) while a value of one represents complete inequality (i.e. entire flux in one instant).
For each solute we generated C-Q plots and fit the log-transformed data with a linear regression (i.e. a power function in un-transformed data) to determine whether they exhibited an enrichment, dilution, or chemostatic response (Godsey et al., 2009). A positive slope indicates enrichment, a near-zero slope indicating relative chemostasis, and a negative slope indicates dilution; a slope of exactly -1 is a special case indicating perfect dilution of a constant flux of solute. We carefully examined the C-Q relationships for any signs of hysteretic behavior at both annual and event time scales, noting the directionality. Clockwise hysteresis indicates relative enrichment of earlier arriving water, while counter-clockwise hysteresis indicates relative enrichment of later arriving water (Evans and Davies, 1998).
The NEON sensor array has only been deployed since fall 2017, providing four years of data at the time of this analysis. To better characterize drivers of inter-annual variability, and better contextualize the results in the context of long term trends, we supplemented this data with historical LTER records of daily average Q (Williams et al., 2021) and weekly grab samples from 2004 through 2014 (Williams 2021). This data was publicly available from the Environmental Data Initiative (EDI) data portal (https://portal.edirepository.org/). To understand the potential for the different sampling frequencies of the historic data to influence estimates of annual flux, we downsampled our sensor-based solute measurements to match the intervals used in the historic sampling (daily average Q and concentration measurements from every Monday at 12:00). Using these values, we re-calculated the estimated annual fluxes of DOC and NO3-N and compared with the estimates made using the 15 minute data.