Salt Dilution Uncertainty and Proposed Metadata Requirements CWRA Conference Vancouver Gabe Sentlinger Mar 7, 2013

Quantifying Uncertainty (Content AZ, Photo GS) Don’t torpedo your measurement! -We all have different methods for SD, we want to identify sources of error/uncertainty and control for them. -Ensure calibration, repeatability, linearity of method Use standard error propagation relation In most cases δM is small (0.1 %) can be in the range of 1-10 % δCF.T is based on calibration procedure and site specific regression if no calibration was done (2-7 %)

Possible RISC Standards: Error Budget So maybe we have an Error Budget that we need to work within for a given RISC Class. Where do we want to spend that budget, where is the low- hanging fruit?

Lowest Fruit: Mixing Error Currently unquantified. Multiple probes, or injections, can help determine appropriate mixing reach for given Q This is a single pulse measured on left and right bank, and thalweg. Standard Deviation ±21%. Left Bank Q = cms Thalweg Q = cms Right Bank Q = cms Average Q = cms Swoffer Q = cms

Impetus for Study: Automated Gauging –How much error is introduced by assuming a CF.T for automated measurements? –Is CF.T a function of EC.T? It shouldn’t be. –Is CF.T a function of Instrument? It shouldn’t be. –Is Background EC.T constant over measurement?

Constant CF.T or Site Specific? This histogram shows that 95% of CF.T values are within 5% of the median value. So if you just used the median CF.T value, only 5% error introduced into Q measurement. Recent improvements to methods may eliminate the site specificity.

Need for fast, accurate, reproducible method –Reduce the amount of salt required for a given flow/ reduce uncertainty associated with salt dilution measurement. –Establish SOP to ensure data quality and traceability, quantify uncertainty, and protect sensitive habitat.

Instrument Accuracy/Resolution –What is the limit of the instrument? –This is a Unidata 6536B with resolution of 0.01 uS/cm but accuracy of 0.5% of the reading. We found quanta of 0.3uS/cm.

Instrument Accuracy/Resolution –Oakton Con110 has better accuracy and resolution at lower ranges, higher SNR, therefore less uncertainty in Q.

Calibration Party! -Salter kitchen scale 5000±1 g ml: $35. -MyWeigh BCS-80 scale 80±0.02kg: $171. -JScale HP-50x 50±.01g: $29.

Calibration Party! -10 ml graduated plastic syringe stated error ±0.01 ml, free(!) at pharmacy ml ± ml, which is 1.4%. - I use 5.02 ± ml, or 1.0%. -No significant temperature effect over range of interest (0-25ºC)

Calibration Party! μl pipettor Diamond Pro : 275$ -measured 997 ± 5ul (0.5%) -did not measure temperature dependence

Calibration Party! -5 ml glass pipette stated error ±0.01 ml, with bulb 15$. -41 measurements, average volume / ml (0.7%).

Calibration Party! ml plastic graduated cylinder stated error ±10ml at 20ºC: 64$ -20 measurements, average volume /-2.1 ml (0.4%). -temperature dependence of 0.1 ml/ºC, 1ml in range from 0-10ºC, less than measurement error.

Calibration Party! -500 ml glass volumetric flask ±0.2ml at 20ºC: 35$ -not calibrated, assumed to be within specs. -Total cost of calibration party, 22$, reduction of total error by approximate 5%.

Need for fast, accurate, reproducible method –Uncertainty should be a trade off between effort/cost and accuracy. –For example, lab glassware is more accurate, but more expensive, fragile, and difficult to use in the field. –Plasticware is less expensive, more rugged, but less accurate and subject to temperature effects. But to what degree, should we worry about it? –Our tests show that there is a temperature effect of 0.1 ml/oC.

Calibration Factor Error (in progress) We’ve (AZ) identified a problem with a pre-mixed standard solution of salt and distilled water. The distilled water in the standard dilutes the total solution; like running in sand you move forward each time, but your reference point moves farther back. I worked out the equations for 5ml injections of 3 std concentrations in 500ml of stream sample. This can be a significant source of error for this method of CF.T derivation, especially at higher background EC.T values.

Calibration Factor Correction (in progress) The bias is positive, CF.T is overestimated; Q is proportional to 1/CF.T so it produces estimates of Q that are lower than true. The correction to the error I’ve worked out (for distilled water solute) to be: Assuming the increase in mass for each injection is only the mass from the added NaCl A more accurate representation of the mass for each standard injection (i is initial in sample, s is std injection; will change subscripts in V2.0)

Draft Metadata (1/2) Metadata required to assess quality (uncertainty) of measurement Useful to have to determine sources of error, better understand the measurement

Draft Metadata (2/2) If uncertainty cannot be assessed, a nominal uncertainty is assumed, ±15-30%.

Estimate of Uncertainty It’s unlikely that all errors would align. It’s good to do repeat measurements, and repeat CF.T measurements, for a given stage to determine the repeatability of the measurement.

~Fini Questions? (Ask Andre, I need a nap)