Reading path: Coastal water observation →
Find the dataset’s conventions first
Before calculating anything, inspect the data dictionary, file notes, and column descriptions for missing-value codes. A blank cell may indicate no observation; a value such as -999, 9999, or a text marker such as NA may serve as a placeholder. The exact convention depends on the dataset, so do not assume that a number is a real reading or a missing marker based only on its size.
For example, suppose a temperature file contains -999 in several rows. If those entries are coded as unavailable measurements, the first cleaning step is to convert them to a true missing state. Leaving them as numbers would pull an average far below the observed temperatures and distort a graph. Check whether the code applies to every column: a placeholder used for temperature might be a legitimate value in another field.
Keep unknowns out of arithmetic
A zero has a clear meaning: the instrument or observer recorded a value of zero in the stated units. Missing means the value is not available. Replacing every blank with zero changes the question being answered. If a sensor reports temperatures of 8 and 10 degrees on either side of a gap, adding zero for the gap lowers the mean even though no zero-temperature observation was made.
Count valid observations alongside summaries. A daily mean based on one measurement is not equivalent in support to a daily mean based on a full set of readings. State how missing entries were handled, and avoid silently dropping rows if the lost observations cluster at a particular time or location. A malfunction during rough weather, for instance, could make the remaining sample unrepresentative.
Make plots show the gap
When a time series has missing intervals, leave a visible break or mark the absent observations. Connecting points across a brief gap may be useful for a visual guide if clearly identified, but it should not imply that intermediate values were measured. Do not plot missing codes at their numeric position; a line plunging to -999 is a data-cleaning error, not a dramatic marine event.
For a spreadsheet workflow, create a separate cleaned column that converts documented missing codes to blank or a recognized null value. Preserve the original file, and keep a note of the conversion rule so the process can be repeated.
Explain what remains unknown
A useful caption names the variable, units, and treatment of unavailable readings. If a long outage affects interpretation, identify its dates rather than hiding it in a general note. The reader can then distinguish a quiet period with measured zeros from a period when the record says nothing.
Before sharing a result, compare the count of original rows with the count of usable observations and inspect a few converted entries. That small audit helps catch accidental substitutions. Missingness is part of the evidence about how a record was collected; preserving it makes conclusions more honest and easier to reproduce.
Have a correction or a source to suggest? Send a note →