Ask what the smallest change an instrument can catch is, and the honest answer is not zero, because every measurement carries some noise that a genuine change has to rise above to be seen. The minimum detectable change is that floor: the smallest shift in the real input that the instrument can reliably tell apart from its own restlessness. This guide defines the minimum detectable change, separates it from accuracy, explains the three things that set it, and shows why it caps the smallest leak, level nudge, or flow change a SCADA system can actually catch, and how it should drive alarm and compression deadband settings.
Minimum Detectable Change in one line: The minimum detectable change is the smallest change in the true input that an instrument can reliably distinguish from its own noise. It is set together by resolution, sensitivity, and the noise floor, and it is different from accuracy, which is about how close a reading is to truth. A change smaller than the minimum detectable change disappears into the noise and cannot be trusted as a real event.
The minimum detectable change answers a specific question: by how much must the real input move before the instrument produces a change in reading that we can confidently say is caused by the process and not by noise? Below that amount, any wiggle in the output could just as easily be the instrument's own electrical noise, quantization, or environmental jitter, so a genuine change of that size is indistinguishable from doing nothing. The minimum detectable change is therefore the practical detection threshold of the measurement, the point below which the instrument is effectively blind to change.
Three properties combine to set it. Resolution is the smallest increment the output can represent at all, for instance the step size of the analogue-to-digital conversion; a change finer than one step simply cannot appear. Sensitivity is how much the output moves for a given change in input, and low sensitivity means a real change produces only a faint output shift. The noise floor is the background level of random variation always present in the signal; a change has to exceed that variation by a clear margin to be recognized rather than mistaken for another noise excursion. The minimum detectable change is set by whichever of these is most limiting, and often by their interaction.
Because noise is random, detection is a matter of confidence rather than a hard line. A change equal to the noise floor is buried; a change several times larger is obvious; in between there is a region where the change is probably real but not certain. Practitioners often express the minimum detectable change as the input shift needed to clear the noise by a stated factor, which is why the same instrument can have a smaller detectable change if you are willing to average over more time or accept a higher chance of a false trigger. Detection is always a trade between how small a change you want to catch and how many false alarms you will tolerate.
It is easy to conflate the minimum detectable change with accuracy, but they answer different questions and can point in opposite directions. Accuracy is about how close a reading is to the true value, an absolute matter of correctness. The minimum detectable change is about how small a movement the instrument can resolve from noise, a matter of discrimination. An instrument can be inaccurate yet excellent at detecting change, because a consistent offset does not stop it from seeing that the input moved; conversely a device can be well calibrated and accurate yet unable to detect a small change because its noise floor swamps it.
A concrete way to see the difference is to imagine a level transmitter that reads a fixed amount high due to a calibration offset. Its accuracy is poor, but if it is quiet and sensitive it may still catch a very small rise in level, because the offset shifts every reading equally and does not hide the change. Now imagine a second transmitter that is perfectly calibrated but noisy; it reports the correct level on average yet cannot tell a small, slow rise from its own jitter. For catching a slow leak or a creeping level, the second instrument's accuracy is little help, because what matters there is the minimum detectable change.
This is why specifying an instrument for change-detection duty means looking past the headline accuracy figure to resolution, sensitivity, and noise behaviour. Leak detection, small-flow measurement, and slow-drift monitoring live or die on the minimum detectable change, not on absolute accuracy. Understanding which quantity a given task actually depends on prevents both under-buying, where a device cannot see the change that matters, and over-buying accuracy that the application never uses.
The minimum detectable change places a hard ceiling on the smallest real event a SCADA system can ever register, because SCADA only sees what the field instrument can resolve. If a pressure transmitter's minimum detectable change corresponds to a certain small pressure drop, then a leak that produces a smaller drop is invisible no matter how clever the software, and a level nudge or flow change below that floor never becomes data at all. Recognizing this floor keeps expectations honest: a monitoring scheme cannot reliably catch changes smaller than the instruments feeding it can distinguish from noise.
This directly shapes two settings that matter for every analogue point: the alarm deadband and the compression deadband. An alarm deadband set finer than the minimum detectable change guarantees nuisance alarms, because the point will cross the threshold repeatedly on noise alone, with no real event behind it. Set the deadband comfortably above the noise floor and genuine changes still trigger while random flutter does not. The compression deadband, which controls how much a value must move before the historian stores a new point, has the mirror problem: set it below the noise and the historian fills with noise; set it too far above the minimum detectable change and small but real events are discarded before they are ever recorded.
For a cloud SCADA platform such as Merobix, gathering data from many remote sites over constrained links, tuning these deadbands against each instrument's minimum detectable change is where sensitivity meets efficiency. The goal at every point is to place the deadband in the window that sits above the noise floor but below the smallest event worth catching, so the platform stores and alarms on real change without drowning in noise or paying to transmit and keep meaningless wiggle. Getting that window right per point is what lets an operator trust that a small level move flagged in the cloud is a genuine event rather than an instrument twitching, and that a quiet trend really means nothing is happening rather than that a real change slipped under the floor.
Resolution is the smallest increment an instrument's output can represent, such as one step of the analogue-to-digital conversion. Minimum detectable change is the smallest real input shift that can be reliably distinguished from noise, and it is usually larger than one resolution step because noise, not just step size, has to be overcome. Resolution is one of the three factors, along with sensitivity and the noise floor, that set the minimum detectable change.
Because SCADA can only register changes the field instrument itself can resolve from its own noise. If a leak produces a pressure or flow change smaller than the instrument's minimum detectable change, that change disappears into the noise floor and never becomes trustworthy data, regardless of how the software is configured. Catching smaller events requires instruments with a lower noise floor, or accepting slower detection through longer averaging.
The alarm deadband should sit comfortably above the minimum detectable change so the point does not cross the threshold on noise alone. A deadband set finer than the noise floor causes nuisance alarms with no real event behind them, while a deadband set well above the smallest meaningful event can hide genuine changes. Choosing the deadband in the window above the noise but below the smallest event worth catching gives reliable alarms without false triggers.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.