I have talked previously about information and uncertainty in hydrometric data. Information refers to how well the data represent the ‘true’ signal whereas uncertainty refers to excursions from the truth that are entrenched in the data. Measuring uncertainty using statistical methods (e.g. Guide to the Expression of Uncertainty) gives us the ability to quantify misinformation but these methods are currently inadequate to even identify the presence of disinformation in data.
Consider the question:
Who formulated the theory of relativity?
a) Albert Einstein
b) Albert Einstien
c) Richard Feynman
d) A scientist
Answer (a) is ‘true’ information. Answer (b) contains misinformation i.e. it is a garbled version of the truth. Answer (c) is disinformation i.e. it is believably wrong with high apparent precision. Answer (d) is also disinformation – in this case it has low fidelity with respect to the ‘truth’ being asked for.
These two examples of disinformation illustrate epistemic error (i.e. an error in knowledge) and procedural error (i.e. an error in the process of seeking the truth). Both forms of disinformation can exist in hydrometric data and both can be difficult to either completely prevent or to detect and mitigate. Furthermore, hydrometric disinformation can result in false inference about other hydrologic processes. There is a high level of trust put in hydrometric data and information voids are filled by the accounting needed to close the water balance.
Hydrometric disinformation is therefore very insidious in that it provides a false understanding of hydrologic systems as opposed to aleatoric errors that merely make it more difficult to develop an understanding of hydrologic processes.
Several examples of hydrometric disinformation are explored in more detail in Hamilton and Moore (2012) but it is worth highlighting how rich the opportunity is for conflation of procedural with epistemic errors. Hydrometric monitoring is often done in adverse conditions. Conditions that are unfavorable for discharge measurements are often also unfavorable for the use of a stage-discharge rating. This results in discharge measurements with disinformation (i.e. low fidelity with respect to the ‘truth’) having high leverage on decisions in the transformation of stage to discharge (i.e. epistemic error).
As in the examples Dan and I provide in Hamilton and Moore (2012) it is easier to identify disinformation with the benefit of hindsight. Retrospective analyses using all of the accumulated information acquired since the data were originally produced can be highly informative. The limiting factor in these analyses is always metadata that either never existed or has since become disassociated from the data.
One small example of missing metadata is documentation about line tagging for sounding measurements. In retrospective analysis it is usually easy to identify whether the sounding weight was sufficient for the depths and velocities measured. It is usually easy to identify if wet line / dry line corrections were applied. However, it is usually impossible to identify if tags were used on the sounding line. If yes, the dry line error has been mitigated leaving a relatively small wet line error as disinformation. If no, then the disinformation in the measurement can be very substantial. It is precisely the highest measurements on record that have the greatest risk of this procedural disinformation and these measurements also have the highest leverage on epistemic error in terms of influence on the shape of the rating curve. Why do standard field notes not have a dedicated field for tag usage? Why are there no standard operating procedures for verification of tag distances above the weight on measurements where they are used? Why do field visit summaries not indicate whether the sounding weight was adequate for the depths and velocities encountered and, if not, what method of mitigation was used and how was the effectiveness of the mitigation verified?