Appendix 1 – Supplementary Methods
a) Chemical analysis of water samples
Dissolved organic carbon (DOC) and total dissolved nitrogen (TDN) concentrations were quantified by the combustion method using a Shimadzu TOC-VCSH analyzer. In each sample run, internal check standards were included at concentrations between 1 and 20 mg L-1 C as potassium hydrogen phthalate and between 0.1 and 1 mg L-1 N as NO3-. Maximum deviations from expected values were 0.1 mg L-1 for C and 0.06 mg L-1 for N. NH4+ and nitrate NO3- were determined by colormetric analysis on filtered samples run through a flow analyzer (Astoria-Pacific). The standard deviation of 50 mg L-1 check standards run ten times during a sample run was 5 mg L-1 for NO3- and 4 mg L-1 for NH4+. Considering two standard curves run at the start and end of each run (0, 5, 10, 50, 100, 300 mg L-1), the maximum deviation of any individual value at the end of run from the expected value was 2 mg L-1 for NO3- and 8 mg L-1 for NH4+. DON was calculated as the difference between TDN and NO3- plus NH4+.
b) Data gaps and equipment positioning
Of the 365 days over which the instrument was installed, measurements were not be collected during the following time periods: 1) between December 9th, 2009 and January 22nd, 2010 where ice formation under low flow conditions prevented measurement, 2) 5 days between November 5th and 10th, 2009 when leaves became lodged over the probe face, 3) 7 days between May 11th and 18th, 2010 where a data logger programming error resulted in data loss, and 4) On July 10th, 2010 where a small event was sampled utilizing an autosampler and the rinsing routine of the autosampler prior to sample collection disturbed debris within the small pool where the FDOM probe was installed, causing attenuation of the FDOM signal. Beyond these 56 days the only other gaps in data collection occurred where no flow was present and water levels dropped below the probe level.
The position of the FDOM probe was shifted slightly (<30cm) in January to reduce the potential impact of reduced mixing at the sensor location with dropping water level, but was moved 40m upstream in April as velocity patterns continued to change, and then slightly (<30cm) in late July to remain submerged after flow temporarily ceased.
c) Methods for correction of FDOM and generation of high-temporal resolution estimates of DOM chemistry
A strong linear relationship was observed between DOC and raw in-situ Cyclops-7 measurements of FDOM (a signal voltage output) collected in the field at times corresponding to sample collection (r2 = 0.85, p <0.0001). However, as compared to check measurements of FDOM made systematically in the lab using a bench top fluorometer, Cyclops-7 measures of FDOM were observed to vary with both temperature and probe position. Given that a critical aspect of this study was to evaluate the utilization of FDOM measures to develop predictive models for DOM chemistry and bioavailability, it was deemed inadequate to use uncorrected in-situ measures of FDOM in further analysis or to simply convert raw measures to quinine sulfate units (Q.S.U.) using a linear relationship generated for a single temperature and using an artificial quinine sulfate substrate (as is the manufacturer recommended practice for most FDOM probes). For this reason the following methods were utilized:
1) Temperature effects were corrected using a simple empirical formula derived by manipulative experiments in the lab. Although temperature corrections were made prior to the publication of Watras and others 2011, we employ a very similar temperature correction method to that which they recommend for correction of CDOM fluorescence sensor measurements. In experiments to define a temperature correction factor, stream water samples of differing DOC concentration were used to systematically identify the impact of changing temperature and initial concentration on probe-based measures of FDOM intensity. Temperature corrected FDOM measures are expressed for a constant temperature (10°C) using a measure we will refer to as specific FDOM and that is analogous to specific conductivity. FDOM measurements were collected continuously during these experiments using the same instrument system used to make FDOM measurements in the field. The probe for the FDOM system was mounted so as to be suspended near the middle of a water sample filling a 1L low transparency plastic beaker. During the experiments FDOM was monitored in six stream water samples with DOC ranging from 1-15 mg L-1 while first decreasing temperature from 20°C to 4°C by indirect cooling in an ice bath and then by increasing temperature from 4°C to 22°C by indirect warming in a warm water. In all cases the relationship between probe-based FDOM measures and temperature was strongly linear with a negative slope (Appendix 1c - Figure 1). This relationship was expected, in that for most fluorescing compounds thermal quenching of fluorescence occurs as temperature increases. After plotting these relationships it was evident that the intercept (FDOM0) varied with starting concentration and that the slope of the relationship also varied with concentration, but could be expressed as a function of intercept (FDOM0 )(Appendix 1c - Figure 1). In this sense, the slope also differs with concentration, but calibration lines for differing FDOM intensity do not overlap within the temperature range observed in the field (Appendix 1c - Figure 1). Because slope can be expressed as a function of FDOM0 (Appendix 1c - Figure 1, Formula 1), Formula 1 and Formula 2 can be integrated to form Formula 3, which allows for calculation of FDOM0 and slope based only on temperature and a raw FDOM measurement (Formulas 1 and 3).
Formula 1: Slope = FDOM0 * -0.1263 + 0.4262
Formula 2: FDOM = Slope * temperature + FDOM0
Formula 3: FDOM0 = (FDOM -0.4262 * temperature) / (1 + temperature * 0.0126)
Using the slope and FDOM0 values from Formula 1 and Formula 3 in Formula 2 provides an FDOM measure for a specific temperature that is analogous to specific conductivity. We used a specific temperature value of 10°C. It should be noted that the value of the coefficients used in these formulas are likely to differ by sensor system and with units used to express uncorrected measures, but the nature of these relationships should be consistent.
Appendix 1c - Figure 1
Appendix 1c - Figure 1 shows (a) the impact of changing temperature on fluorescent dissolved organic matter readings (FDOM) by the Turner Cyclops-7 FDOM probe for varying starting concentrations of dissolved organic matter and shows (b) the relationship between the slope and intercept of the FDOM versus temperature relationships for the differing concentrations measured. FDOM is expressed in relative fluorescence units (RFU) as a measurement system specific value.
2) Inner-filtering effects were identified to also influence FDOM measurements taken in the field. We attempted to develop an empirical inner-filter correction function for in-situ FDOM measurements by utilizing the relationship between absorbance at 254nm (UV254) and FDOM intensity measured on the Cyclops-7 as is recommended in Downing and others 2012, but the nature of this relationship was identified to differ seasonally for our dataset (Appendix 1c, Figure 2). The seasonal nature of this trend was verified by comparing bench top measured FDOM to UV254 of grab samples. For each sample, FDOM measures were made on a bench top fluorometer (Varian Cary Eclipse) as the intensity for the single wavelength combination of excitation 370nm, emission 470nm . This wavelength combination corresponds most closely with those used by the Turner Designs probe. Bench top fluorometer readings were converted to R.U. prior to analysis using the area under a Milli-Q blank scatter peak at excitation 350nm. A Beckman DU 520 spectrophotometer was used to measure absorbance spectra and these spectra were utilized to correct benchtop FDOM for minor inner filter effects (Walker and others 2009). As expected, correction of benchtop FDOM measurement for minor inner filter effects creates a linear relationship between FDOM and UV254; however, seasonal differences in the slope of this relationship remain evident (Appendix 1c, Figure 2). The relationship between FDOM measured in the stream and UV254 exhibits increased impact of inner-filter effect with increasing concentration, but the same general pattern of increasing slope in the relationship between FDOM and UV254 from fall through winter, snowmelt, and summer was evident when comparing UV254 to in-situ FDOM measurements (Appendix 1c, Figure 2). Together, these patterns seem to indicate that this discrepancy in response may relate to a shift in DOM character. Minor changes in instrument position occurred between seasons along with changing water level, but given the nature of the relationship between FDOM and UV254 in both the lab and field we assume that changing DOM composition and not position drive the seasonal differences observed. Season specific calibrations were generated to account for the attenuation of light by inner filter effects. These seasons were based on dominant environmental changes: fall was defined as the period from leaf fall through to ice formation (October 30th to December 12th, 2009), winter as the period from ice formation through to the onset of snowmelt (December 12th to March 5th, 2009/2010), snowmelt was defined by the onset of the snowmelt freshet and ended with its recession (March 5th to April 25th, 2010), and summer was defined as spring through to leaf fall (April 25th to October 30th, 2010). Because we completed bench top measurements of FDOM, rather than creating correction factors for light attenuation utilizing an empirical relationship with absorbance or DOC measurements (Downing and others 2012) we calibrated specific FDOM directly to laboratory FDOM (Appendix 1c - Figure 3). The form of this relationship reflects the increasing level of light attenuation at higher FDOM as was illustrated by Downing and others (2012). For this reason the slope of the regression between probe and bench top based measures does increase slightly with concentration (bench top data is SQRT transformed in Appendix 1c - Figure 3). Interference from turbidity was assumed to be negligible because turbidity measures were generally low and were not systematically higher with higher DOC concentration (data not shown, maximum total suspended solids observed was 18 mg L-1, average 5.9 mg L-1).
Appendix 1c – Figure 2
Appendix 1c – Figure 2 – The upper figure panel shows fluorescent dissolved organic matter (FDOM) as measured on the Varian Cary Eclipse bench top fluorometer plotted against absorbance at 254 nm (UV254) as measured on a bechtop spectrophotometer. The lower figure panel shows FDOM measured by the Turner Cyclops-7 FDOM probe corrected for temperature to 10 degrees Celsius and plotted against UV254 as measured in the lab.
Appendix 1c - Figure 3
Appendix 1c - Figure 3 – Fluorescent dissolved organic matter (FDOM) measured by the Turner Cyclops-7 FDOM probe corrected for temperature to 10 degrees Celsius and plotted against measures of FDOM made using the Varian Cary Eclipse bench top fluorometer. For both systems peak emission was measured at 470nm with excitation at 370nm. In-stream FDOM intensity is expressed as relative fluorescence units (RFU).
3) After correction of FDOM measures for the impact of temperature and inner filter effects, linear relationships with DOC, DON, and BDOC were identified by plotting concentrations measured for discrete samples against measurements that were recorded in the field at the time of sample collection. For BDOC, distribution of concentrations was non-normal so transformation was required to approximate the linear relationship. Using linear regression, high temporal resolution time series for each DOM chemical parameter were generated over the period of sample collection (October 30th 2009 to October 30th 2010) (Appendix 1c – Figure 4).
Appendix 1c – Figure 4
Appendix 1c - Figure 4 – DOC, DON, and BDOC as predicted by FDOM. FDOM has been corrected for temperature effects to 10 degrees Celsius, corrected for seasonal inner filter effects, and calibrated to Raman Units based on equivalent measurement on a Varian Cary Eclipse bench top fluorometer. Dotted gray lines indicate the 95% prediction intervals for each relationship.
d) Eliminating redundancy of environmental variables, transformations to meet the assumption of normality, and model selection using Akaike weights
Mean water temperatures and soil temperatures over the course of each event were highly correlated (r = 0.98), as were the 15-minute interval measurements of each (r = 0.97), reflecting the strong connectivity between the soil water and stream water watershed components. Given the wide range of conditions examined, average air temperature and water temperature for each event were also strongly correlated (r = 0.91). As a result, only water temperature was included as a predictor in multiple factor regression because this measure was made in closer spatial proximity to the sampling point. For the 1, 3, and 5 day antecedent volumes of water exported, a strong correlation existed with each other and starting Q (r values ranging from 0.95 to 0.98). Because these measures all provide an indication of antecedent moisture conditions, only starting Q was retained as an indicator of moisture condition because hydrological modeling for the Bigelow Brook has shown that baseflow Q relates linearly to catchment-water storage (Xu and others 2012). To meet the assumptions of normality, the input data were transformed prior to multiple factor regression. Total volume of water exported during the event, volume of water exported during the preceding storm, magnitude, magnitude of the preceding event, event length, and length of rising limb were natural log transformed. Starting Q and time since last event were square root transformed.
Akaike Information Criterion (AICc) was used to select those models for flow weighted DOC concentration (DOCFW) with a high probability of best fit. Then DAIC was calculated as a measure of relative likelihood of a model being the best model and Akaike weights (wi) were calculated to provide an approximation of the probability of a model being the best fit among those models considered. For additional information on this statistical method see Westphal and others (2003), but generally, models with DAIC ≤ 2 have substantial support as candidate models and wi can be considered as a measure of probability of a given model being the best of the set of models with values ranging from 0 to 1.