Here in Ashland, Oregon, today (Dec 20, 2020) is the shortest day length of the year at 9:05:45. But the winter solstice isn't until 2:02 AM PST tomorrow (Dec 21, 2020). Tomorrow's day length is a whopping three seconds longer at 9:05:48. Why is tomorrow's day length greater than today? It would seem to me that tomorrow's sunrise and sunset are slightly closer to the magic moment of winter solstice and should therefore account for a slightly shorter day.
The other possibility is https://sunrise-sunset.org is inaccurate.
- An app that I use, LunaSolCal, show the length of day on Dec 20, 21, and 22 in Ashland OR are 9:05:45, 9:05:44, 9:05:49.
- The website https://www.timeanddate.com does not have Ashland OR in its database, so I used nearby Medford OR. The length of day on Dec 20, 21, and 22 are 9:04:50, 9:04:49, and 9:04:53.
Although those two sources have a difference of 1 minute in the length of the day, the pattern is correct and expected: the shortest length of daylight is on the day closest to the winter solstice. Therefore, I think sunrise-sunset.org is off by one day with its calculations.
I was very surprised when I first heard this and it took a little while to understand it.
The simple answer is to do with the M in GMT. This is "mean" in its sense of average. So, clock noon to the next clock noon is a fixed amount of time: the average day. If you used a sundial, then you would find that sundial noon to sundial noon was not always the same length. It is this variation in the length of the sundial day that causes the surprising non-coincidence of shortest day, latest sunrise, and earliest sunset. If you use a sundial instead of a modern clock then you would not see this surprising behaviour.
As Martin says, look up the Equation of time for why the length of the day varies and how it does so.
This is an interesting question which opens up wide vistas! Look up the Equation of Time on Wikipedia for a start.
What is your source? I can't find the sunrise and sunset times in Ashland. The ESRL site for Ashland gives noon to the second but sunrise and sunset only to the minute.
At present the days are 24 hours and 30 seconds long, so each noon is 30 seconds later, by the clock, than the last one. This "later each day" applies to all solar events including sunset and sunrise, and this is the reason why the sunsets are already getting later each day while the sunrises don't start getting earlier until early January.
What puzzles me is that this does not by itself account for the discrepancy you are seeing in actual sunrise-to-sunset length, which is why I asked about the source.