Heavy rain is a topic which lacks the intense drama of, say, melting ice. Nevertheless, changes in heavy rain frequency are important because they are cited as evidence of AGW and because, unlike Arctic ice, increased flooding could have global social impact.
This particular post concerns a tiny part of the subject – something I term “precipitation time-of-observation bias” (“P-TOB” for short).
P-TOB is not mentioned in climate science literature, so far as I can tell. This contrasts with P-TOB’s well-known cousin (temperature time of bias, “T-TOB”) which plays an important role in temperature reconstruction. P-TOB, unlike its cousin, goes without note.
The lack of literature reference certainly makes me pause and wonder if my reasoning is wrongheaded and P-TOB actually does not exist. Or, maybe my search has missed the explanation of how P-TOB is removed or avoided in studies, or maybe on balance any effect is too small to matter – I don’t know. But, nothing conjectured, nothing learned, and public errors can be excellent teachers. So, here goes.
Here’s the basic idea: a lot of intense rain occurs as thundershowers, especially on summer afternoons. If the 24-hr rainfall observation occurs during an afternoon thundershower then that thundershower may be split in two and recorded as two moderate rains rather than one intense rain. A record-keeping problem develops if the observation time is later changed to the unstormy early morning period. Such a change would mean that afternoon thundershowers are no longer split and are instead recorded as one intense rain rather than two moderate rains.
This shift in observation time (P-TOB) may make it appear that thundershowers have become more intense while in reality they have not changed.
Here is an illustration which may help:
If the daily rainfall total is taken at “A” ( see the red line) then the total afternoon storm is captured into one day of the record, as an intense event. However, if the observation is taken at “B” (the blue line) then the storm is split between two days and recorded as two moderate events.
If the time of observation changes from B to A and that change is not recognized when records are examined then one may think that there has been a shift towards intense precipitation events. However, that apparent shift may actually be an illusion created by the change in observation time.
That’s the basic P-TOB idea.
To reinforce the rainy-afternoon idea here is an example of summertime hourly rainfall distribution. I used Asheville, NC (home of the NCDC) as the example because that hourly data is free and readily available on the internet:
Can the P-TOB concept be illustrated via data? Yes. For that effort I again used Asheville NC 1998-99, but this time I used all months (even though P-TOB may be mainly a summertime affair). I looked at the daily rainfall distributions for observation times of midnight, 6 AM, noon, 3 PM and 6 PM. Below is an example, comparing 6 AM and 6 PM observation times and showing the heavy-rain parts of the distribution curves. I can provide other combinations if anyone so wishes:
The 6 AM observation time for 1998-99 tends to capture the afternoon intense storms as single events while the 6 PM observation tends to split those intense storms into two less-intense storms.
Suppose I split the example’s entire distribution into quantiles of 5%, for ease of viewing. Here is the result:
In this presentation P-TOB shows a distinct pattern, with an apparent sharp increase in the most-intense precipitation (right side) offset by smaller net declines throughout the remainder of the distribution. The net, of course, is zero. If one was unaware of P-TOB then the spike on the right might be reported as a real phenomena rather than as a data illusion.
Is this odd change in distribution seen in any of the studies? I checked Karl et al’s 1998 study
on changes in intense precipitation in the US and noted this plot for the southeastern US (the location of Asheville). The column graph shows the change in precipitation distribution over the course of the 20’th Century, with the most-intense on the right side:
Is the similarity between the two a coincidence? It indeed could be. Or, maybe P-TOB affected Karl et al’s results. I don’t know. However I do know that I have difficulty conceiving of a physical reason for the distribution found in Karl et al.
Below are a few other questions and graphs.
** Have precipitation times of observation changed? Yes, at least in North Carolina. Below is a plot of the changes in observation times of the twenty-five North Carolina USHCN stations which report daily values:
As the plot indicates, there has been a notable shift from afternoon to morning precipitation observations in North Carolina since 1950 (I didn’t review pre-1950 records). If P-TOB exists then this shift in observation time would tend to result in the appearance of more-intense daily rainfall events in the records.
** What does the intensity time series look like for those North Carolina stations? The following plot takes the 25 stations as a group and counts the days of intense rain (defined as 2 or greater inches in a day) per year:
At face value the plot shows a trend towards more days of intense rain in North Carolina.
Suppose we break the group in two and first look at just the stations which changed from afternoon to morning observations. Their result is here:
It’s a sharp pattern of increasing days of intense rain.
Now how about those stations which had no change in observation time:
It’s essentially unchanged – no increase in the number of days of intense rain at stations which had no change in observation time. That’s quite a contrast versus the prior plot.
These illustrations are small samples which are examined in only one manner and may not be representative of the larger body of data. No conjecture about the actual extent of P-TOB is possible without a fuller examination. At this point my only goal is to float the concept of P-TOB and its possible consequences, and provide a modest midsummer diversion while the ice melts, or not.
Footnote: the free Asheville 1998-99 hourly data can be found here: