Check timing and measurement quality

An observation time records when a sensor measured conditions. A reception time records when the receiving system obtained the message. An availability time records when a processed value became accessible. SCD-1 relays information within a system where these times can differ; calling every record real-time without defining the delay hides an important limitation.

INPE’s SINDA description updated in June 2025 describes processing and availability within thirty minutes after reception. The starting point matters: this is not a statement that every measurement reaches a user within thirty minutes of being taken. Waiting for a suitable pass or retry can occur before reception, and a user’s later download is another separate event.

In an original example, a station samples at 12:00, reception occurs at 12:18 and the record becomes available at 12:28. The sampling-to-availability delay is 28 minutes; reception-to-availability processing takes ten minutes. Both numbers are useful, but they answer different questions. Substituting one for the other could make a monitoring system appear faster than it is.

Quality checks are also separate from speed. Suppose a fictional calibration converts a sensor count to centimetres using level = 0.5 × count + 10. A count of 240 represents 130 centimetres. If the reference datum changes, retain that change in the station history. A sudden step may come from an instrument adjustment rather than a sudden physical rise in the river.

Different clocks answer different questions

Consider expected reports at 13:00, 13:10 and 13:20. The first and third contain readings of 1.2 and 1.4 metres; the middle report is absent. The mean of the two received readings is 1.3 metres. Inserting zero for the missing report would invent an observation and lower the result to about 0.87 metres. Even the valid-sample mean does not reveal what happened during the gap.

A review should keep missingness visible, check station maintenance and compare neighbouring evidence where appropriate. Interpolation may be useful for a clearly stated analysis, but an estimate must not silently replace a measurement. Preserving the raw report, interpreted value and quality flag lets later users revisit a choice without losing the original evidence.