A little more on very long term time series

I wrote earlier about a visit to Rothamsted Research, to talk with them about some of their very long term time series of agricultural research data (since 1843... digitised since 1991). Asking around, the prevailing wisdom seems to be to break the time series data when the nature of the data changes. Keep those time series, un-touched. Then build your overall time-series by a set of transformations on the original datasets, where the actual transformations are well-documented.

Of course if (as is perhaps the nature of such agricultural experiments) the nature of the data changes pretty well every year, then you have to keep a series of one-year data snapshots. And that sounds pretty much like what Rothamsted's batch "sheets" are doing.

Meanwhile, I'm continuing to look for something a bit more authoritative than "prevailing wisdom"! Someone has promised me a reference to some Norwegian economic or price time series kept since the 16th century, so I'm hopeful! Any hints from readers welcome...