Reading path: Coastal water observation →
Organize before judging
A spreadsheet can make an unexpected reading easier to find if each row contains a timestamp, site, variable, value, unit, and any instrument or field note. Sort by site and time so neighboring observations are visible together. Check for duplicated rows, inconsistent decimal separators, unit changes, and timestamps that have been imported as text; these can resemble unusual measurements.
Keep a copy of the original file before making changes. Add a separate review column for a suspected outlier and explain why it was flagged. An unusual value is a candidate for investigation, not automatically an error. A real environmental event can produce a sharp and important extreme.
Plot the series to see context
A scatter or line plot with time on the horizontal axis can show whether a point is isolated, part of a rapid transition, or one end of a longer pattern. Use clear units and avoid connecting across long gaps as if measurements were continuous. A second plot by site can help reveal whether a value is unusual only relative to one station or across the set.
Suppose a coastal temperature record contains 9.2°C, 9.3°C, 15.8°C, then 9.4°C at regular intervals. The spike is visually distinctive, but the plot does not tell whether it came from a sensor glitch, a temporary plume, or a data-entry error. Inspect the raw row, nearby sensors, and deployment notes before deciding.
Investigate with independent clues
Check field notes for handling, exposure, instrument repositioning, or maintenance at the time. Review quality flags and the sensor's resolution. Compare with another instrument only after checking depth, location, sampling frequency, and units. If the value sits outside a plausible range, treat that as a reason to investigate rather than a universal rejection test.
A typo may be suggested by an extra digit or a unit mismatch, while a sudden but plausible shift may match a real environmental change. Record evidence for the decision. If the cause remains uncertain, retain the value and mark its status rather than silently removing it from the dataset.
Make cleanup traceable
When a calculation excludes a reading, state the criterion, affected row or time, and supporting reason. Keep original values alongside any corrected or excluded version. Do not replace a missing observation with zero, since zero is a measurement and can alter averages and plots. If a correction is justified, label it as such and preserve the source value for comparison.
This workflow makes the spreadsheet a review aid rather than a verdict machine. A clear note can say that a value was flagged for review because it was isolated and inconsistent with instrument notes, while also acknowledging that its physical cause was unresolved. Readers can then assess the evidence and revisit the decision.
Have a correction or a source to suggest? Send a note →