I agree, but not because of the very slight drift you’ll experience over time.
Millis uses an unsigned long variable type, which means that it resets to zero after around 49.7 days. This makes it very difficult to use in a scenario like yours. RTC is a much better approach in my opinion.
You will of course get a 25 hour and a 23 hour day day once per year, when the clocks change from summertime to wintertime, but I don’t think that’s going to cause you too many issues.
You may also want to think about the usefulness of simply knowing rainfall in the past 24 hours. Sure, it’s something that the weather forecasters use, but you may find it more useful to know how much rain fell per hour, then keep a rolling total of the last 24 hours.
I actually monitor rainfall in the past minute, 60 minutes and 24 hours. In the case of the 60 minute and 24 figures, these are rolling periods rather than periods defined by the clock.
There’s a more detailed discussion on the issue in this thread:
Pete.