Performance Verification Statement for In-Situ Troll 9000 Rugged Dissolved Oxygen Sensor.

Abstract

The Alliance for Coastal Technology (ACT) conducted a sensor verification study of in situ dissolved oxygen sensors during 2015-2016 to characterize performance measures of accuracy and reliability in a series of controlled laboratory studies and field mooring tests in diverse coastal environments. The verification including several months of Laboratory testing along with three field deployments covering freshwater, estuarine, and oceanic environments. Laboratory tests of accuracy, precision, response time, and stability were conducted at Moss Landing Marine Lab. A series of nine accuracy and precision tests were conducted at three fixed salinity levels (0, 10, 35) at each of three fixed temperatures (5, 15, 30 oC). A laboratory based stability test was conducted over 56 days using deionized water to examine performance consistency without active biofouling. A response test was conducted to examine equilibration times across an oxygen gradient of 8mg/L at a constant temperature of 15 oC. Three field-mooring tests were conducted to examine the ability of test instruments to consistently track natural changes in dissolved oxygen over extended deployments of 12-16 weeks. Deployments were conducted at: (1) Lake Superior, Houghton, MI from 9Jan – 22Apr, (2) Chesapeake Bay, Solomons, MD from 20May – 5Aug, and (3) Kaneohe Bay, Kaneohe, HI from 24Sep – 21Jan. Instrument performance was evaluated against reference samples collected and analyzed on site by ACT staff using Winkler titrations following the methods of Carignan et.al. 1998. A total of 725 reference samples were collected during the laboratory tests and between 118 – 142 reference samples were collected for each mooring test. This document presents the performance results of In Situ Troll 9000 rugged dissolved oxygen (RDO) sensor using optical luminescence technology. Instrument accuracy and precision for the Troll 9000 RDO was tested under nine combinations of temperature and salinity over a range of DO concentrations from 10% to 120% of saturation. The means of the difference between the Troll 9000 and reference measurement ranged from -0.289 to 0.173 mg/L. There was a small difference in the mean offset for the 4 oC trials (mean = 0.05 mg/L) versus the 15 or 30 oC trials (means = -0.05 and -0.06 mg/L, respectively). A small response differences was also noted across salinity levels with a mean offset of 0.05 mg/L for the 0 salinity trials compared to -0.07 and -0.04 mg/L for the 10 and 35 salinity trials, respectively. A global linear regression of the instrument versus reference measurements for all trials combined (n=356; r2 = 0.99; p<0.0001) produced a slope of 1.005 and intercept of -0.075. The absolute precision, estimated as the standard deviation (s.d.) around the mean, ranged from 0.002 – 0.013 mg/L across trials with an overall average of 0.004 mg/L. Relative precision, estimated as the coefficient of variation (CV% = (s.d./mean)x100), ranged from 0.021 – 0.268 percent across trials with an overall average of 0.062%. Instrument accuracy was assessed under a 56 day lab stability test in a deionized water bath cycling temperature and ambient DO saturation on a daily basis. The overall mean difference between instrument and reference measurements was -0.040 (s.d. = 0.517) mg/L for 75 comparisons. There was no significant trend (linear regression r2 = 0.009, p=0.41) in accuracy over time that would indicate performance drift; however the magnitude of offset clearly increased after approximately 30 days. A functional response time test was conducted by examining instrument response when rapidly transitioning between adjacent high (9.6 mg/L) and low (2.0 mg/L) DO water baths, maintained commonly at 15 oC. The calculated τ90 was 52 s during high to low transitions and 48 s for low to high transitions covering the 8 mg/L DO range. At Houghton, MI a field deployment test was conducted under the ice over 104 days with a mean temperature and salinity of 0.7 oC and 0.01. The Troll 9000 RDO operated successfully throughout the entire 15week deployment and generated 9859 observations based on its 15 minute sampling interval for a data completion result of 100%. The ambient DO range captured by the reference samples was 10.249 to 14.007 mg/L compared to the slightly broader dynamic range of 9.33 to 14.71 mg/L recorded by the Troll 9000. The average and standard deviation of the measurement difference over the total deployment was 0.680 ± 0.072 mg/L with a total range of 0.422 to 0.940 mg/L. The drift rate of instrument offset, estimated by linear regression (r2=0.29; p<0.001), was -0.001 mg/L/d. This rate would include any biofouling effects as well as any electronic or calibration drift. A linear regression of the instrument versus reference measurements over the first month (r2 = 0.98; p<0.0001) produced a slope of 0.93 and intercept of 1.61, indicating an initial calibration offset. At Chesapeake Biological Lab, a field deployment test was conducted over 78 days with a mean temperature and salinity of 25.6 oC and 10.9. The Troll 9000 generated 1879 acceptable measurements (based on ± 2 mg/L from nearest reference sample) from a possible 3639 observations based on its 30 minute sampling interval for a data completion result of 52%. The ambient DO reported by reference samples was 4.370 to 10.858 mg/L compared to the broader dynamic range of 2.01 to 12.18 mg/L measured by the Troll 9000. The average and standard deviation of the difference between instrument and reference measurements for the deployment was 0.550 ±0.409 mg/L, with the total range of differences between -0.420 to 1.068 mg/L. The drift rate of instrument offset for the subset of data was -0.146 mg/L/d (r2=0.814). This rate would include any biofouling effects as well as any electronic or calibration drift. However it is likely that data included after 6/20 reflect a failing instrument and not a normal drift response. A linear regression of the instrument versus reference measurements for the first month of the deployment (r2 = 0.699, p<0.001) produced a slope of 1.234 and intercept of -1.384. At Kaneohe Bay, HI a field deployment test was conducted over 121 days with a mean temperature and salinity of 25.8 and 33.4 oC. The Troll quit operating on after 26 days and generated 579 out of 2826 possible observations based on its hourly sampling interval for a data completion result of 21%. The average and standard deviation of the differences between accepted instrument measurements and reference readings (n=33 of possible n=129) were -0.032 ± 0.402mg/L, with a total range in the differences of -1.458 to 0.482 mg/L. No calculation of a drift rate is included given the short operating interval and high variability which suggested an immediate problem within the instrument. A linear regression of instrument versus reference measurements for the subset data (r2 = 0.96, p<0.001) had a slope of 1.117 and intercept of -0.692. Overall, the Troll 9000 showed good linearity across all three salinity ranges including fresh, brackish, and oceanic water, covering a dissolved oxygen range if between 4 to 14 mg/L. A global linear regression of the composited data (r2 = 0.963; p<0.001)) had a slope of 1.086 and intercept of -0.351. The Troll 9000 was evaluated in a profiling field test in the Great Lakes at two separate locations in order to experience transitions from surface waters into both normoxic and hypoxic hypolimnion. In Muskegon Lake, the temperature ranged from 21.0 oC at the surface to 13.5 oC in the hypolimnion, with corresponding DO concentrations of 7.8 and 2.8 mg/L, respectively. In chigan, the temperature ranged from 21.0 oC at the surface to 4.1 oC in the hypolimnion, with corresponding DO concentrations of 8.6 and 12.6 mg/L, respectively. Two profiling trials were conducted at each location. The first trial involved equilibrating test instruments at the surface (3m) for ten minutes and then collecting three Niskin bottle samples at one minute intervals. Following the third sample, the rosette was quickly profiled into the hypolimnion were samples were collected immediately upon arrival and then each minute for the next 6 minutes. The second trial was performed in the reverse direction. For Muskegon Lake, the Troll 9000 overestimated DO when transitioning rapidly from the hypolimnetic equilibration into the warmer and high DO surface water. Conversely when equilibrate in the surface water the Troll RDO underestimated DO levels when rapidly transitioned into the colder, low DO hypolimnion. The range in measurement differences between instrument and reference was 0.04 to 0.41 mg/L for cast 2 and -0.35 to 0.23 mg/L for cast 3. (Note: cast 1 was aborted and redone as cast 3). For Lake Michigan, the Troll 9000 underestimated DO when transitioning rapidly from surface equilibration into a colder, high DO hypolimnion. Conversely when equilibrate in the hypolimnion the Troll RDO overestimated DO levels when rapidly transitioned into the warmer, lower DO surface waters. The range in measurement differences between instrument and reference was -0.59 to 0.07 mg/L for cast 1 and -0.11 to 0.54 mg/L for cast 2. In both cases the Troll RDO appeared to approach equilibrium by the seventh minute and the offsets declined to around 0.10 – 0.15 mg/L.