We designed two studies to examine how different environmental factors affected metabolism and decomposition rates in Lake Superior Tributary streams, located in the vicinity of the Keweenaw Peninsula in the Upper Peninsula of Michigan. The first, which we will refer to as “the longitudinal study”, examined the variation in ER, HR, AR, and decomposition along the Pilgrim River with sites distributed from the headwaters to just upstream of where the river joined the Keweenaw Waterway. We hypothesized that ER (AR + HR) rates would be higher in downstream sites than in headwater sites, due to an increase in rates of GPP and AR with increasing light availability at downstream sites. We predicted no difference in HR between sites, as heterotrophic activity should not be affected by light availability. We also hypothesized that decomposition rates would be lower in downstream sites where the availability is leaf litter is proportionally lower compared to autochthonous carbon. The second study, which we will refer to as “the DOC gradient study”, examined the variation and environmental predictors of metabolism and decomposition among four streams with different DOC concentrations. We hypothesized that ER rates would not be affected by DOC concentrations, due to the offsetting responses of HR and AR. We hypothesized that higher DOC concentrations would stimulate HR but also result in more browning of the water (de Wit et al. 2016) which would limit light availability, causing AR to decrease. We also hypothesized that rates of ER, AR, HR, and decomposition would be greater at sites with warmer water temperatures.
For the longitudinal study, five sites were chosen along the Pilgrim River, located in Houghton County, Michigan. The sites ranged from 1 km upstream of the mouth of the river where it enters the Keweenaw Waterway ("River Km 1”), another two sites located 6 km (“River Km 6”, also the location of USGS gauge 04043016) and 14 km ("River Km 14”) upstream from the mouth of the river, up to two sites on headwater tributaries (“River Km 15” and “River Km 16”). The DOC gradient study included the Pilgrim “River Km 1” site as well as sites in three other rivers: the Trap Rock River, the Silver River, and the Tobacco River. The Trap Rock River site was located 6 km upstream from the river mouth entering Torch Lake, at USGS gauge 04043050 near Lake Linden, Michigan. The Silver River site was located 4 km upstream of the mouth of the river entering Huron Bay of Lake Superior, at USGS gauge 04043150 near L’Anse, Michigan. The Tobacco River site was located near Gay, Michigan. Environmental characteristics, such as water chemistry, canopy cover, and stream width were measured once every two weeks from May 2020 until October 2020.
Environmental Characteristics: The methods used were consistent across both studies. To characterize the physical and chemical conditions at each study site, we measured canopy cover and concentrations of nutrients in the water. Canopy cover was measured using a densiometer. Stream water was collected from each site and filtered using Millipore 0.45 μm nitrocellulose membrane filters into acid washed plastic Nalgene bottles for measurement of dissolved water chemistry. Samples were frozen until analysis was performed for nitrate (NO3-), ammonium (NH4+), total dissolved phosphorus (TDP), dissolved organic carbon (DOC), total dissolved nitrogen (TDN), and soluble reactive phosphorus (SRP). Water chemistry samples and canopy cover measurements were collected once a month for the duration of the longitudinal study, and once every two weeks for the duration of the DOC gradient study. The DOC and TDN concentrations were measured using a Shimadzu TOC-VCSN analyzer with a total N module TNM-1 (Shimadzu Scientific Instruments, Columbia, Maryland). The NH4+ concentrations were measured using a fluorometric procedure (Holmes et al. 1991; Taylor et al. 2007) on a Turner Aquafluor (Turner Designs, Palo Alto California). The TDP concentrations were measured using acid-persulfate digestion followed by molybdenum-antimony colorimetric determination methods with a Thermo Scientific 10 s UV–Vis spectrophotometer (Ameel et al 1993; APHA 2005; Nydahl 1978; Valderrama 1981). SRP was analyzed on a SEAL AQ2 discrete analyzer (SEAL Analytical, Mequon, Wisconsin) based on USEPA method 365.1 revision 2.0 (USEPA 1993a) and APHA method 4500- P F (APHA 2005). NO3- was also analyzed on a SEAL AQ2 discrete analyzer (SEAL Analytical, Mequon, Wisconsin) based on USEPA method 353.2 revision 2.0 (USEPA 1993b) and APHA method 4500 NO3- (APHA 2005).
Metabolism Modeling: Dissolved oxygen and temperature were measured at 10-minute intervals using a miniDOT dissolved oxygen (DO) sensor (PME Instruments, Vista CA) at each study site (see Online Supplementary Materials Figures 1-4 for the full datasets). Sensors were deployed at each site from 14 May until 6 September, 2019 for the longitudinal study and 12 June to 20 October, 2020 for the DOC gradient study, except at the Tobacco River where the sensor was deployed on 29 July, 2020. Discharge was measured manually using a Marsh-McBirney Flo-Mate Model 2000 Portable Flowmeter (Hach, Loveland, CO) once a month at the Pilgrim River sites during the longitudinal study. For the DOC gradient study, discharge was measured every two weeks at the Tobacco River, and obtained from the USGS gauges at the remaining sites.
Dissolved oxygen, temperature, stream depth, estimated from stream width and discharge, and discharge were used to estimate GPP and ER with Bayesian inverse modeling. We used the streamMetabolizer package for R (Appling et al. 2018, R Core Team 2020) to evaluate the single-station metabolism equation. PAR was estimated from sampling date and time, site longitude, and site latitude using streamMetabolizer (Appling et al. 2018). We fit the model using the default prior probabilities of GPP ~ N(μ = 3.1, SD = 6), ER ~ N(μ = -7.1, SD = 7.1) and k600 ~ N(μ = log(12), SD = 1.32), where k600 is a temperature-normalized version of K. Using inverse modeling guided by prior probability distributions, the model uses O2, water temperature, depth, light, to determine estimates of GPP, ER, and K for each full day of measurement. The model structure assumes that ER is constant throughout the day, and nighttime change in O2 is attributable to only ER and K. As ER represents the consumption of O¬2, we report the values as negative numbers, where greater |ER| represents greater respiration.
Flow velocity was estimated with Q using the discharge data. As for Tobacco River, Silver River, Trap Rock River, and the Pilgrim River in 2020, all four streams had no major discontinuities upstream of the study reach, so a single station assumption was appropriate.
Once the GPP and ER modeling was complete, we performed linear regressions in R (R Core Team 2020) comparing daily mean ER and k600 to look for equifinality in the model. Equifinality is where many different combinations of estimated values for GPP, ER, and k600 can fit different O2 records, or large values of all three are just as likely to fit the data as ecologically realistic values (Appling et al. 2018). In order to examine the accuracy of the k600¬ estimates, we examined whether the model converged on reasonable metabolism estimates using the using the rstan package (R Core Team 2020, R Stan Development Team 2020). None of the DOC gradient study sites (the Pilgrim River km 1 site in 2020, Tobacco River, Silver River, and the Trap Rock River) exhibited equifinality, but both of the headwater sites in the longitudinal gradient study (Pilgrim River km 15 and km 16) did exhibit equifinality, and therefore we were only able to successfully model metabolism for 3 of the 5 longitudinal study sites (Pilgrim River km 1, km 6 and km 14).
Autotrophic and Heterotrophic Respiration - A quantile regression approach was used following the procedure of Hall and Beaulieu (2013) to estimate rates of AR using the quantreg package in R (Koenker 2020, R Core Team 2020). This approach uses the 90th percentile slope of quantile regression between ER and GPP multiplied by daily GPP to calculate AR (Hall and Beaulieu 2013, Equation 2), as AR is hypothesized to be the minimum amount of ER on any day above the base HR (Hall and Beaulieu 2013). Where ARf is the slope of the 0.9 quantile regression between ER and GPP, representing the fraction of primary production that is immediately respired by autotrophs. As total ER should be comprised of AR and HR, HR can be estimated by subtracting AR from ER. Because this is a statistical approach to estimate these different respiration rates, there are strict assumptions that must be met, which we assessed following the recommendations by Hall and Beaulieu (2013). No estimates were used where the confidence interval of the quantile regression slope was greater than 0.4. The correlation between HR and GPP was also examined, and we did not use any sites with a correlation of 0.3 or higher, as that was shown to decrease or increase the ARf estimates by 0.15, depending on the sign of the correlation. Also, any sites that had low variation in GPP were found to have unrealistic estimations of AR rates (as in Hall and Beaulieu 2013), so these sites were excluded from further analyses. Based on these criteria, AR and HR were estimated for the Silver River, Pilgrim River km 6, and Pilgrim River km 1 sites and could not be estimated for the Pilgrim River km 14, Tobacco River, and Trap Rock River sites.
Decomposition - Cotton strip assays were deployed at each site following the procedure outlined in Tiegs et al. (2013). Five cotton strips were deployed at each site and were incubated for 27 - 33 days. Upon collection, the strips were washed with 90% ethanol, then dried at 40 °C and placed in a desiccator to await tensile strength measurement. Lower tensile strength, or the amount of pulling force a material can receive until it breaks, should correspond to greater decomposition of the cotton fibers. The tensile strength of the incubated strips, as well as control strips, were measured by pulling at a rate of 2 cm min-1 to measure maximum tensile strength using an Instron 4206 load frame (Instrom, Norwood, MA) with an MTS Renew controller and 45.35 kg load cells (MTS Systems, Eden Prairie, MN). Decomposition rates were calculated following Mancuso et al. (2022). We calculated decomposition rates using both the number of days the strips were incubated (referred to as kd, representing decomposition rate per day) or using the degree days, calculated similarly to Mancuso et al. (2022) (referred to as kdd, representing decomposition rate per degree day).
References cited:
Ameel, J. J., R. P. Axler, and C. J. Owen. 1993. Persulfate digestion for determination of total nitrogen and phosphorus in low nutrient water. American Environmental Laboratory 5:2-11.
American Public Health Administration (APHA). 2005. Standard Methods for the Examination of Water and Wastewater. 21st Edition. American Public Health Association/American Water Works Association/Water Environment Federation, Washington, DC.
Appling, A.P., R. O. Hall Jr., C. B. Yackulic, and M. Arroita M. 2018. Overcoming Equifinality: Leveraging Long Time Series for Stream Metabolism Estimation. Journal of Geophysical Research: Biogeosciences 123:624–645.
de Wit, H. A., S. Valinia, G. A. Weyhenmeyer, M. N. Futter, P. Kortelainen, K. Austnes, D. O. Hessen, A. Räike, H. Laudon, and J. Vuorenmaa. 2016. Current browning of surface waters will be further promoted by wetter climate. Environmental Science and Technology Letters 3:430–435.
Hall, R.O. Jr., and J. J. Beaulieu. 2013. Estimating autotrophic respiration in streams using daily metabolism data. Freshwater Science 32:507–516.
Holmes, R. M., A. Aminot, R. Kérouel, B. A. Hooker, and B. J. Peterson. 1999. A simple and precise method for measuring ammonium in marine and freshwater ecosystems. Canadian Journal of Fisheries and Aquatic Sciences 56:1801-1808.
Koenker, R. 2011. quantreg: quantile regression. R package version 4.71. R Project for Statistical Computing, Vienna, Austria.
Mancuso, J., E. Messick, and S. D. Tiegs. 2022. Parsing spatial and temporal variation in stream ecosystem functioning. Ecosphere 13:e4202.
Nyadhl, F. 1978. On the peroxodisulfate oxidation of total nitrogen in waters to nitrate. Water Research 12:1123-1130.
R Core Team. 2020. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/.
Taylor, B. W., C. F. Keep, R. O. Hall Jr., B. J. Koch, L. M. Tronstad, A. S. Flecker, and A. J. Ulseth. 2007. Improving the fluorometric ammonium method: Matrix effects, background fluorescence, and standard additions. Journal of the North American Benthological Society 26:167–177.
Tiegs, S. D., J. E. Clapcott, N. A. Griffiths, and A. J. Boulton. 2013. A standardized cotton-strip assay for measuring organic-matter decomposition in streams. Ecological Indicators 32:131–139.
USEPA. 1993a. Method 365.1, Revision 2.0: Determination of Phosphorus by SemiAutomated Colorimetry.
USEPA. 1993b. Method 353.2, Revision 2.0: Determination of Nitrate-Nitrite Nitrogen by Automated Colorimetry.
Valderrama, J. C. 1981. The simultaneous analysis of total nitrogen and total phosphorus in natural waters. Marine Chemistry 10:109–122.
