there are conceivable cases were it seems natural to not keep the "base data" around.
Your example does not demonstrate this point. By taking the temperature readings in degrees C, you, in fact, are taking the "base data". Of course, you also should record the equipment that you are using to take that data. The height of mercury in your thermometer can be determined by the the brand of thermometer (including the Manufacturer and Model #), and the temperature reading given. Most scientific peer-reviewed research articles are very specific about the equipment being used.
"By taking the temperature readings in degrees C, you, in fact, are taking the "base data"."
I don't see how? It is a matter of how you define "the base of the data". My point is that there is a calculation on the measurement, and only the result of the calculation is recorded. Hence the base data is discarded.
We don't know what calculations were done on the climate date in question here. But it seems at least possible that the calculation is also trivial (for climate researchers anyway, could be some standard procedure), just as to us converting heights to degrees is trivial.
Recording the type of thermometer is another matter - likewise the climate researchers will probably record where their measurements were taken, which implies the equipment used.
Edit: of course if the calculations are reversible is another issue. I see now that is the point you made with the thermometer (calculate height from make and degrees). Well, it was just an example, and we don't even know yet if the calculations in question here are not reversible. I suppose if they calculated an average they are not. But I also think there are probably many cases where a measurement is just taking an average (probably even the height of quicksilver is an average in a way...).
we don't even know yet if the calculations in question here are not reversible.
??? If the calculations are reversible, there would be no argument. Everyone would know the base data.
I think you are kind of pushing it with your quicksilver example; I don't see how taking a single measurement is "averaging". If you mean to say that there is some built-in variance - well, that's true. But the fact is the variance itself can be established by other scientists if they know the model of thermometer that is being used. This is the essence of repeatability!
Now, imagine that a scientist had many temperature readings, and then they did some kind of complex calculation which involved regression analysis, statistical modeling, etc. These are the types of calculations that could easily have errors. Without the initial recordings, the results are unverifiable. That doesn't make them necessarily useless, but I wouldn't lean on those kinds of results.
Your example does not demonstrate this point. By taking the temperature readings in degrees C, you, in fact, are taking the "base data". Of course, you also should record the equipment that you are using to take that data. The height of mercury in your thermometer can be determined by the the brand of thermometer (including the Manufacturer and Model #), and the temperature reading given. Most scientific peer-reviewed research articles are very specific about the equipment being used.