Measuring Turbidity: New Revelations
High-resolution measurements of settling formazin using a SediMeter with a high measurement rate reveals small-scale turbidity cells. An implication of this discovery is that Lindorm in the past has overestimated the noise in the sensor; much of the apparent noise may have been real data, an effect of packets of water with different turbidity moving about.
Although the SediMeter was invented for measuring changes in bottom level, by virtue of its design it is also a vertical array of optical backscatter detectors, i.e., 36 individual turbidimeters 10 mm apart. To make it useful for turbidity measurements we have now added an export function of turbidity data. The other step we are taking is to revisit the calibration, to see if we can increase the precision.
We started by looking for and quantifying the various factors that contribute to errors during calibration (some of these are also relevant once the instrument is being used). One of the factors we looked at is how one should best stir the formazin calibration standard, and how it behaves after stirring ceases (i.e., how often must one stir?).
This experiment had been done with the original prototype of the SediMeter, with 1000 FTU concentration, 30 minute measurement interval. It had seemed that the concentration remained rather constant for about an hour, wherafter it sank uniformly so that a front of clear water moved downwards.>/p>
When we now repeated the experiment with a new SediMeter, having a longer probe, much higher turbidity resolution, measuring every minute, and using a 2000 FTU suspension, the result was unexpected:
One measurement per minute, stirring with a plunger ceased just before first measurement. For turbidity scale see bottom of page.
The clear water front that moves downward after one to two hours is consistent with the earlier observation. However, with the higher measurement frequency, and the improved accuracy thanks to calibration and 16-bit resolution rather than 8-bit, the presence of a pattern of turbidity variations is revealed almost from the start.
One can still see some noise, random fluctuations in value in the horizontal direction (horizontal lines on the other hand would be due to bubbles or dirt on the sensor). The experiment was repeated a few days later, to confirm the result, after again re-calibrating the SediMeter. This time it was set to measure once every 20 seconds. The result was even more unexpected:
The first observation is that the general pattern is similar to the previous experiment. A darker (less turbid) packet of water is rising from the bottom towards the surface, starting about 15 minutes in. A turbid packet starts from near the surface and sinks down, meeting the darker field after about 50 minutes. But here is the fascinating novelty: This turbid region in the graph has got intricate details. (Note that the time-scale is different, since the experiment was interrupted sooner.)
The turbidity in that region fluctuates by as much as 10% within a minute. It is clearly not noise but data. However, by resampling the data to simulate one measurement per minute (below; it has also been re-scaled for comparison with the first graph), we can see that at the once per minute measurement frequency the fluctuations will mostly appear as noise. We can also see that although the formazin settled in a similar way in the two experiments, they were not identical. Presumably the process takes place by upwelling and downwelling developing in different parts of the tube (5 cm in diameter), why the exact details will depend on chance. But the general pattern the first hour, with the knee on the dark strike, was repeated each time (about a half dozen undocumented tests were also made).Turbidity scale in FBU:
Conclusion: The accuracy of the SediMeter sensor when used without a holder tube has been estimated as 6 FBU, refering to the average of the 6 top detectors - which is what is reported in center graph on the Data window, and exported with the Save button. To arrive at that value, a higher uncertainty of the individual detectors was decreased by a formula that takes into account the averaging. However, these new results tentatively suggest that the real accuracy may in fact be better than 6 FBU for an individual detector. When the data has been analyzed we will update this statement, so for now the old specification is left as is.