This is a rather common situation in electronics which is called capacitor soakage. It is not bizarre at all in some situations. For first order approximations it is seldom used. It does come up in sample and hold circuits where the sample time is short and the hold time is very long. Depending on capacitor type and the times involved it can be a non-issue or it can ruin your month.

Speculation:
Perhaps the bug in averaging module manifests on a particular situation in the data (perhaps missing records), but disappears with later datasets with more complete data. Or … the bug has been fixed but averaging has not been redone as far back as March 2008.

Since GHCN continues to update past records as new data arrives, at least some of the past monthly averages have to be redone each month. I don’t know whether this done only for stations flagged with new data or globally, and so far, I haven’t seen this level of detail in the NCDC documentation.

I sorted the 800 stations that had data for all 10 months and checked for Mar-Apr and Sep-Oct possible carryovers (i.e. 0 difference in these months). These three northern Canadian stations are definitely wrong in April (includes Resolute previously found by Steve).

(Reminder: Divide by 10 to get actual temp averages, i.e. -286 = -28.6)

StationID__ Name_________ Jan Feb Mar Apr May Jun Jul Aug Sep Oct
403 71081 HALL BEACH,N. -286 -366 -308 -308 -36 16 59 38 -2 -67
403 71926 RESOLUTE,N.W -295 -345 -313 -313 -30 52 134 104 23 -38
403 71924 BAKER LAKE,N -317 -351 -322 -322 -73 21 53 23 -50 -108

These two are possible carryover errors in Oct., but these would need to be confirmed by actual calculation of the avarage from station data

309 84628 LIMA-CALLAO/A 225 230 237 203 174 177 186 177 175 175
207 42933 AKOLA (India)… 228 227 292 325 347 311 283 273 276 276

After that there are about a dozen “zero” difference records that are likely chance occurrences in the tropics, e.g.:
149 63862 DODOMA (Tanz.) 238 224 220 220 215 200 198 208 220 244

I haven’t checked all the months, but I did notice that the Finland July-August carryover error appeared to be gone.

Tentative conclusions:
- The “carryover” error appears to occur in clusters within countries.
- There are still some such errors in the 2008 record, but most likely a small number of them, and the effect is probably insubstantial for the most part (but of course needs to be fixed)
- Any fix in averaging should be applied to all 2008 data, and previous years should perhaps be examined (I’ll look at 2007 when I have a chance)
- the sparse northern stations may have an upward effect on the April anomaly when corrected.

Steve and Anthony were also mentioned the other day on the Mark Levin show, based on the Times article. They were referenced indirectly on the Glenn Beck show as well.

Are they using the invalid data GISS got from them, or are these figures based on corrected data??

If the former, are there invalid data from the SH as well as NH causing both hemispheres to appear spuriously warm? I thought the GISS/CRU problem was concentrated in the NH.