When you add up the sources of error in a measurement, it is tempting to think each input contributes in proportion to how uncertain it is. But that is not how it works, because some inputs affect the result far more strongly than others. The sensitivity coefficient is the factor that captures that difference: it says how much the final result changes for a change in each input, and so it decides how heavily each input's uncertainty counts. This guide explains what a sensitivity coefficient is, how it comes from the measurement equation as a partial derivative, and why it explains the everyday observation that a small differential-pressure error can dominate a flow uncertainty budget while a temperature error barely registers.
Sensitivity Coefficient in one line: A sensitivity coefficient is the factor that tells you how much the result of a measurement changes when one of its inputs changes, and it is used to weight that input's uncertainty when the contributions are combined into a total. Mathematically it is the partial derivative of the output quantity with respect to that input, so it comes straight from the measurement equation. In a flow uncertainty budget the sensitivity coefficients are why inputs the flow depends on strongly, such as differential pressure, contribute far more to the total than inputs the flow depends on weakly, such as temperature.
An uncertainty budget lists the sources of error in a measurement and combines them into a single figure for the result. The mistake a newcomer often makes is to assume the biggest uncertainty on the list is the biggest contributor to the total. That is not necessarily true, because two things determine how much an input matters: how uncertain the input itself is, and how strongly the result responds to a change in that input. The second of those is the sensitivity coefficient, and it is what turns a raw input uncertainty into a contribution to the result.
The contribution of any input to the result's uncertainty is its own standard uncertainty multiplied by its sensitivity coefficient. So an input can be measured quite precisely and still dominate the budget if its sensitivity coefficient is large, and an input can be quite uncertain yet barely matter if its sensitivity coefficient is small. The sensitivity coefficient is, in effect, a weighting factor: it scales each input's uncertainty according to the leverage that input has over the final answer. Only after applying these weights does it make sense to compare contributions and see which sources truly drive the total.
Thinking in terms of these weights changes how you read a budget. Instead of chasing the input with the largest raw uncertainty, you look for the input with the largest contribution, meaning the largest product of uncertainty and sensitivity. Sometimes those are the same input and sometimes they are not. This is precisely the breakdown that general uncertainty-budget explanations tend to skip: they show the inputs and the total but do not make clear why one modest-looking input can be the one that governs the whole result. The answer is always its sensitivity coefficient.
The sensitivity coefficient is not a guess or a rule of thumb; it comes directly from the equation that defines the measurement. If the result is computed from several inputs by some formula, then the sensitivity coefficient for a given input is the partial derivative of the result with respect to that input - the rate at which the result changes as that one input changes while the others are held fixed. This is the mechanism at the core of the law of propagation of uncertainty: each input's uncertainty is propagated to the result through the partial derivative of the model with respect to that input.
Because it comes from the measurement equation, the sensitivity coefficient reflects the actual physics and mathematics of how the quantity is derived. If the result depends on an input only weakly, the partial derivative is small and so is the sensitivity coefficient; if the result depends on an input steeply, the partial derivative is large. The units work out too: a sensitivity coefficient carries whatever units convert a change in the input into a change in the output, so multiplying it by the input's uncertainty gives a contribution in the units of the result, ready to be combined with the others.
An important consequence is that a sensitivity coefficient can depend on the operating point, because the slope of a nonlinear model is not the same everywhere. In a relationship where the output goes with the square root of an input, for instance, the sensitivity to that input is larger at low values and smaller at high ones, so the same absolute input error contributes differently depending on where the measurement is operating. This is why sensitivity coefficients are, strictly, evaluated at the conditions of the measurement, and why an uncertainty budget is a snapshot valid around a particular operating point rather than a universal constant.
The clearest illustration lives in differential-pressure flow measurement, the orifice-meter case that so much field instrumentation is built on. The flow rate is derived from several measured inputs - the differential pressure across the orifice, the fluid density or the pressure and temperature that set it, and geometric and coefficient terms. In the governing relationship the flow rate goes with the square root of the differential pressure, and this square-root dependence is exactly what the sensitivity coefficient captures. It gives the differential-pressure input a substantial leverage over the computed flow.
Now compare the inputs. Because flow follows the square root of differential pressure, a given percentage error in the differential-pressure reading translates into a smaller but still significant percentage error in flow, and differential-pressure transmitters, especially near the low end of their range, can carry a meaningful uncertainty. Temperature, by contrast, enters the flow calculation only weakly - it affects flow through its effect on density, and a small absolute temperature error changes the density, and therefore the flow, by a very small fraction. The sensitivity coefficient for temperature is small, so even a temperature error that sounds large in degrees contributes very little to the flow uncertainty. This is why practitioners say a small differential-pressure error can dominate the budget while a temperature error barely matters: the two inputs have very different sensitivity coefficients.
This has direct value in field operations and SCADA-based measurement. If you know from the sensitivity coefficients that differential pressure dominates the flow uncertainty, you know where to spend effort: on a well-ranged, well-maintained differential-pressure transmitter and on avoiding operation at the very low end of its range, rather than on chasing tiny improvements in temperature accuracy that the budget shows cannot help. A cloud SCADA platform that logs the flow inputs lets an engineer see which measurements are operating where their sensitivity is highest, and target maintenance accordingly. In short, the sensitivity coefficient is the tool that turns an uncertainty budget from a list into a guide for action, showing not just how big the total is but which input to improve to shrink it.
A sensitivity coefficient is the factor that says how much the measured result changes when one input changes, and it is used to weight that input's uncertainty when the contributions are combined. It is the partial derivative of the output with respect to that input, taken from the measurement equation. Each input's contribution to the total is its own uncertainty multiplied by its sensitivity coefficient, so the coefficient decides how heavily the input counts.
In differential-pressure flow measurement the flow rate goes with the square root of the differential pressure, so differential pressure has a large sensitivity coefficient and strong leverage over the computed flow. Temperature, by contrast, affects flow only weakly through density, so its sensitivity coefficient is small. As a result a modest differential-pressure error contributes far more to the flow uncertainty than a temperature error of similar apparent size.
Yes, when the measurement model is nonlinear the sensitivity coefficient depends on the operating point, because it is the slope of the model and that slope varies. In a square-root relationship, for example, the sensitivity to the input is larger at low values and smaller at high ones. This is why sensitivity coefficients are evaluated at the actual conditions of the measurement and why an uncertainty budget is a snapshot around a particular operating point.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.