Hi Fellow Bloggers,
John Cook, CEO of ADMI is providing us with a look at lessons learned in his water quality research. The work will be presented in a more concise, palatable series. The series will be broken down into 5 main themes:
- Reconciling Finished Water with Distribution System Water
- The critical importance of sampling frequency in understanding behaviors
- Multi-site monitoring to reduce variability in event detection
- Using chaos theory to monitor difficult behaviors
- Building process models using real-time historical data
This series will be delivered in multiple parts occurring bi-monthly. We hope you enjoy, and as always please feel free to comment.
Everyone intuitively knows that what changes in source water and finished water quality at the water treatment plant affects distribution system water quality. But unless the data is monitored in real-time at both locations, that relationship and its complexity are not completely appreciated. In addition, real-time monitoring of water quality is frequently used as part of an element of inferential event detection systems. As anyone who has analyzed the data knows, the data exhibits “apparent” random variability due to many forcing functions, from flowing fire hydrants, tank discharge, pump cycling and pressure transients. This can make event detection all the more challenging. Now, of course, there are explanations as to why variability is present, hence the use of the term “apparent” randomness. But, it has been our experience that much of this randomness can be explained by evaluating the behavior of source water and chemical dosing changes at the water treatment plant.
Let me illustrate with two examples: 1) the first has to do with apparent random behavior of distribution system turbidity; 2) the second concerns the variability of total chlorine residual on the distribution system. Let’s look at the first graph of turbidity from the water treatment plant and from a location several miles from the water treatment plant. As the treatment plant is a member of the Partnership for Safe Water, the turbidity levels are consistently below 0.1 NTU. Not so for the distribution system turbidity as we see below:
Are the two charts related? Not apparently so, but let’s suppose that we attempt to correlate finished water quality in an effort to explain the consistently low plant turbidity using multivariate nonlinear methods for optimization. Here is the result once other parameters from the WTP are factored into the correlation. We see that the randomness can be explained to a great extent (R2 = 0.71):
Let’s look at one more example of “apparent” randomness—this in the total chlorine at the storage tank site. Factoring in other parameters from the WTP, including alkalinity, color and the blending of source of supplies, produces an explanation of 68% of the variability. We have seen the incredible complexity of distribution systems as they behave as giant bio-chemical reactors. But much of the water quality behavior can be explained to reduce “apparent” randomness.