Sensor Measurements
We estimated whole-stream metabolism at the reach scale using the one-station, open water method (Odum 1956; Hall 2016) using field-calibrated YSI 6920 V2 multiparameter sondes (Yellow Springs Instrumentation, Yellow Springs, OH, USA). Sonde locations (Figure 1) were determined by landowner permissions. Each sonde recorded dissolved oxygen, turbidity, specific conductance, and water temperature at 15-minute intervals, except at site 6 where a turbidity sensor failed. During the year-long deployment from October 2016 through mid-October 2017, sondes were calibrated in the field using water-saturated air approximately every two weeks. Turbidity and specific conductance probes were calibrated in the field when standard calibration criteria were not met.
Light intensity (LUX) was recorded every 15 minutes with HOBO temperature and light intensity loggers (UA-002-64; Onset, Bourne, MA, USA) mounted above the stream surface adjacent to the sondes. Paired LUX and photosynthetically active radiation (PAR) readings from a HOBO PAR sensor over a 30-day period in the summer were used to create a linear relationship (PAR= 0.00847*LUX, r2 = 0.94, p less than 0.001, i.e. (Long and others 2012)) that was used to convert all light intensity measurements to PAR for the study.
Barometric pressure was measured near Site 2 using a pressure transducer (HOBO U20-001-04 Water Level Logger; Onset, Bourne, MA, USA) until March 2017. After this date, we used hourly barometric pressure data for the Pocatello airport weather station obtained from the NOAA National Climate Data Center.
Water level was measured at 15-minute intervals using pressure transducers (HOBO U20-001-04 Water Level Logger; Onset, Bourne, MA, USA) placed adjacent to each sonde. Manual discharge measurements using either a handheld Acoustic Doppler Velocimeter (FlowTracker; SonTek, San Diego, CA) during wadeable conditions or an Acoustic Doppler Current Profiler (M9, SonTek, San Diego, CA) during higher-flow conditions were used to create stage-discharge relationships for each reach (Guilinger 2017). We did not make direct measures of water level at all study sites after March 2017 because water-level loggers were removed from all sites except Site 6, which is a USGS gaging station. We used discharge at site 6, as well as dummy variables for snowmelt and the seasonal flow recession, to estimate discharge at the other sites. Models were developed for each site using data from March-Oct 2016 and were used to estimate discharge at all sites from February 1 to October 13, 2017. Predicted discharge was compared to measured flow data for February 2017 at each site to quantify model error. These estimates of error are likely larger for February than the March-October period because flows are higher during this month than the estimated flow period. R2 for these relationships ranged from 0.73 to 0.98. Discharge-velocity relationships were determined from manual discharge measurements at each site and were used to estimate mean velocity for each day during the study period.
Much of Marsh Creek is either channelized or naturally incised, simplifying channel geometry to a rectangular prism with consistent wetted width. We used 1-meter LiDAR to measure 20 channel widths over a 200-m reach above each sonde location. Daily reach-averaged stream depth was calculated as Discharge / (Velocity x Reach-Averaged Wetted Width). The assumption of a rectangular channel morphology was reasonable for most sites but may lead to overestimates of depths at sites 6 and 8.
To derive the longitudinal profile and channel slope (Figure 1), elevation was extracted every 5 meters along the channel centerline of Marsh Creek from a 1-meter airborne LiDAR digital elevation model. Channel slope, width, and stream power were averaged over the mean 50% turnover length (see below for turnover length calculations) for each site.
Periodic Sampling
Water samples were collected monthly at each site along Marsh Creek and analyzed for total phosphorus (TP), total dissolved nitrogen (TDN), nitrate (NO3 ), ammonium (NH4+), soluble reactive phosphorus (SRP), and dissolved organic carbon (DOC). Water samples for dissolved solutes were filtered through ashed Whatman GF/F filters within 24 hours of collection and frozen until analysis. TP samples were frozen immediately until analysis. Samples for NO3-, NH4+, and SRP were analyzed on a Lachat QuikChem FIA system (Hach, Loveland, CO) at the Environmental Analytical Laboratory at Brigham Young University (detection limit = 0.02 mg/L, precision within 10 %). Samples for TDN and DOC were analyzed on a Shimadzu TOC/TN analyzer also at Brigham Young University (detection limit = 0.07 mg/L, precision within 5 % and detection limit = 0.2 mg/L, precision within 10 % respectively). TP samples were digested using a persulfate digestion protocol and analyzed as SRP as above.
Sestonic chlorophyll a was measured monthly concurrent with water samples from May 2017 - September 2017. Following an extraction in 90 % methanol, samples were centrifuged, and the supernatant was analyzed for chlorophyll a concentration using a spectrophotometer at the Idaho State University's Center for Ecological Research and Education. Ash-free dry mass was also measured using the sestonic water samples to quantify particulate organic matter in the water column (Steinman and others 2007).
Instream growth of macroalgae and macrophytes was estimated as percent cover for 11 transects every 10 m for 100 m upstream of the sonde using the Braun-Blanquet method (Bowden and others 2007). Surveys were done monthly during the growing season from May 2017- September 2017.
Metabolism Modeling
We used inverse modeling to estimate GPP, ER, and K600 from observed dissolved oxygen time series using ‘streamMetabolizer' package in R (Appling and others 2018a, 2018b). Within streamMetabolizer, we used the model formulation: b_Kb_oipi_tr_plrckm.stan. This is a Bayesian hierarchical state space model that includes both process and observation errors. The model estimates K600 using partial pooling based on mean daily discharge; this approach minimized challenges associated with equifinality and inaccurate parameter estimates. Briefly, this approach leverages the relatively long time series and pools K600 toward a linear relationship between K600 and mean daily discharge for each site (see Appling and others 2018a for details) Without pooling K600 estimates, ER and K600 estimates were significantly correlated; using the partial pooling approach resolved this problem. In our final dataset, ER and K were not correlated within any sites (see Supplemental materials Figure S4 for plots of ER-K600 relationships for each site).
Days with unrealistic estimates of GPP and ER (negative GPP and positive ER) were removed from the final dataset (N=80 days with negative GPP and 17 days with positive ER) and were not used in any subsequent analyses. We also removed 67 days with unrealistically low estimates of K for a given discharge. Some sites had many more instances of unrealistic K estimates that affected site standard deviations (e.g., the coefficient of variation for K/discharge among sites ranged from 0.3 to 3), so we were unable to apply a consistent cut-off (e.g., using a z-score of greater than 2) across sites. Instead, we identified outlier days by visually inspecting timeseries of K, discharge, and discharge/K. This resulted in a total of 1085 valid days among all sites (Table 1).
To evaluate the spatial independence of metabolism estimates among sites, we estimated 50% and 80% turnover lengths of O2 using modeled K600 and velocity calculated from discharge-velocity relationships. Using a 50% turnover length (0.7KO2/V), there were 648 instances of non-independence between adjacent sites (60% of modeled days). Using the more conservative 80% turnover length (1.61KO2/V), there were 892 instances of non-independence between adjacent sites, or 82% of our modeled days. Using an 80% turnover length, Site 2 overlapped with Site 6, but not with any upstream sites. Sites 6, 8, 9, and 10 overlapped for nearly all modeled days. Site 15 was most independent from Site 9 (38% overlapping days). Overlap varied among adjacent site pairs, with overlap on most days for Sites 6, 8, and 9, but less frequent overlap with adjacent upstream sites for Sites 2 and 10.
