Reading path: Coastal water observation →
Inspect timestamps before values
A time series is a sequence of measurements associated with times. To find a possible sensor gap, first learn the expected sampling interval from metadata or the instrument log. If a sensor is expected to report every ten minutes, compare neighboring timestamps: 10:00, 10:10, 10:20 is regular, while a jump from 10:20 to 11:10 indicates missing entries in that span.
Make the gap visible rather than silently filling it. A spreadsheet can sort timestamps chronologically, and a simple plot can show where the line breaks. Do not mistake a change in display format for a real gap: timestamps may be rounded, stored in a different time zone, or affected by a daylight-saving clock change. Check the time convention first.
Look for suspicious repetition
A long run of identical temperatures can be a clue, especially if the water was expected to vary or the instrument has limited precision. It is not conclusive. A stable environment can produce similar readings, and values rounded to the nearest tenth may repeat even when the underlying temperature changes slightly. Record the length and timing of the run, then compare it with the sensor's resolution and nearby observations.
For example, a dozen identical readings during calm overnight conditions deserve a different interpretation from the same repeated value continuing through a documented sensor deployment or battery change. If another nearby instrument shows variation, that raises a question to investigate, but differences in depth and location may explain some of the contrast.
Check discontinuities carefully
An abrupt jump or drop may signal a sensor interruption, a reset, a calibration change, a moved instrument, or a real environmental event. Plot the series with timestamps and inspect the points immediately before and after the break. Then consult deployment notes, maintenance records, and any quality flags. If the transition coincides with a documented instrument swap, treat the two segments as potentially distinct until their comparability is established.
Consider a record that moves from 8.4°C to 11.7°C within one sample interval. The size of the change makes it worth checking, but it does not identify the cause. A second sensor, weather context, or field note may clarify whether a real change is plausible. Avoid automatically deleting the point or smoothing across it.
Mark uncertainty in the analysis
Create a transparent working copy that labels suspected gaps, repeated runs, and discontinuities while preserving the original values. A gap should remain missing, not become zero; zero is a real numeric reading and can distort summaries or plots. If you later exclude a suspect segment, document the rule and retain the raw record for comparison.
This approach helps distinguish an observation from an inference. In a notebook, state the expected interval, the dates affected, the checks performed, and what remains uncertain. That makes a temperature series easier to interpret and gives the next reader a clear route for revisiting the decision.
Have a correction or a source to suggest? Send a note →