PYTHON · NUMPY

How fast a sensor forgets

Why does one sensor produce several different calibration intervals?

Put measured drift rates against a decision threshold and compute how long a reading stays usable.

01Challenge

You are asked how often a soil mesh needs recalibrating. The vendor says annually. Test that against published drift rates and the actual decision the readings feed, then give a number you can defend.

Predict firstWhy does one sensor produce several different calibration intervals?

02Model

Drift is not a defect. It is the normal behaviour of a sensor left in the ground, and it is measured. Ion selective pH sensors drift by 0.3 to 0.5 units per year. Resistive moisture sensors need recalibration every six to twelve months. Salinity above 2 dS/m cuts resistive sensor accuracy by 30 to 50 percent, which is not drift at all but a standing offset that arrives with the soil. None of these numbers is alarming on its own. They become interesting only when you put them next to the decision the reading feeds, because a sensor is usable for exactly as long as its error stays smaller than the difference it is being asked to resolve. That is why there is no universal calibration interval. There is a calibration interval per decision.

03Guided practice

Model drift as a linear envelope from the day of calibration. The reading is usable while the envelope is narrower than the threshold the decision needs. Solve for the time at which the two are equal, once per decision, and notice how far apart the answers are.

04Feedback

PASS WHEN You produced three different answers from one sensor, which is the correct result. A calibration interval belongs to a decision, not to a device.

If your run fails, check:

Drift is in units per year and the thresholds are in units. Divide, then convert to months.
Do not average the three thresholds. The point of the exercise is that they disagree.
If every answer came out the same, check that you used each decision's own threshold inside the loop.

05Retrieve & space

Drift rates for field soil sensors are published and are not small.
A reading is usable while its error stays under the difference the decision must resolve.
There is no universal calibration interval, only an interval per decision.

06Mastery & project

A vendor quotes one calibration interval for a sensor sold into twelve different applications. Say what that number can and cannot mean.

Pick a decision made on sensor data in a domain you know. Find or estimate the drift rate of the sensor feeding it and the difference the decision must resolve, then state the interval. Mark every input you had to estimate rather than find.

← The wrongly known fraction Agreement is not accuracy →

PYTHON · NUMPY

How fast a sensor forgets

Published drift rates are given. Compute the usable life of a pH reading against three different decisions.

You read

the arrays and values already in scope

You change

the code you write in each cell

Fixed

the dataset and the checks that grade you