When Did It Rain?

Dates and times in the PACRAIN database

How should rainfall be recorded? If the rainfall accumulation for a certain observation period is desired it may not be obvious that the observation for the following period is required. It is tempting to attribute a rainfall accumulation to the period for which it is valid. Thus, an hourly rain gauge reading made at 1400 might be recorded as part of the 1300 observation. This is not desirable for real-time observations because it introduces a delay. In the above example, the 1300 observation would not be complete until the 1400 rain gauge reading was made. For daily observations the delay would be even more significant.

The PACRAIN database is not real-time, so the delay described above is not an issue. However, there are other issues that must be considered. The majority of PACRAIN observations are daily, but this refers to a 24-hour period, not a calendar day. This is because rain gauge observations are typically not made at midnight, so the accumulation is spread over two calendar days. It is important to note that without secondary information it is impossible to determine the distribution of rainfall within an observation period. For example, if an observation is made at noon it cannot be assumed that half of the measured rainfall occurred on the day of observation and half on the previous day.

Perhaps it makes sense attribute the accumulation to the calendar day which encompasses the majority of the accumulation period (but not necessarily the majority of the rainfall)? If rainfall measurements are taken every day at 0800 then two-thirds of the observation interval is on the calendar day prior to the observation. If the observation is made at 1600 every day the majority of the observation period falls on the day of observation. Consider two observation sites in close proximity, the first one with an observation time of 0800 and the second one with an observation time of 1600. Now assume that it rains at both sites between 1800 and 2200--the same rainfall event will be attributed to a different day at each site. This inconsistency is far from ideal.

For regional or global applications the date of observation becomes even more problematic because the vast majority of daily rainfall observations are made according the local time, not a standardized synoptic schedule. Date of observation alone is especially irrelevant for the PACRAIN domain, which not only has a large longitudinal extent but is bisected by the Date Line. It is obviously essential, then, to consider not only the date but the time of observation for rainfall data.

In the PACRAIN database a complete timestamp (date and time) is stored for each 24-hour observation. However, limitations in the legacy data format mean that time of observation is only available via the prototype interface. The timestamps for Meteo France (FR) data are UTC; all other timestamps are local time but will be converted to UTC in the near future. Time zone information is available in the site catalog. The timestamp for each observation indicates the beginning of the accumulation period.

5 May 2004