Although the formatting of the SST datasets needs to be completely freshened up, once again, before commenting, I commend the SST collaters for honoring their data by ensuring the preservation of comprehensive metadata – as opposed to their cousins at CRU and GISS. Unfortunately, there don’t seem to be any statistical analyses of SST measurements – by this, I mean, where the authors analyse the actual bias of changing measurement systems and provenance. (Although Thomson et al [Nature 2007] challenged earlier bucket adjustments, they didn’t do the sort of patient data analysis that the field cries out for.
The statistical problems are by no means easy. As I mentioned the other day, I’ve collated data from a few gridcells in the Pacific (one pair at Honiara near Bougainville and one pair at Hawaii.) Today I’ll illustrate the nature of the problem with a graph that took me a long time to articulate. [Click for a larger version]. I’ll explain below.
From the ICOADS data for gridcell 17N, 157W, I calculated the random effect for each month (sort of a monthly anomaly adjustment) and subtracted that from the simple average. I first excluded all SST values below 15 deg C (including many negative values) and all SST values above 35 deg C (including a torrid 96.3 deg C measurement.) The result is shown in grey together with a smooth in black. Overlaid on this (red) is the ERSST3 smooth for the gridcell.
The ICOADS information here has tables describing the various decks. I aggregated the decks (a pretty rough cut) to ship, buoy, NODC, US DOD (autodin) with a separate class for “US navy” – important here. I also did a grouping of the SST Code classes (SI) into bucket, inlet and sensor. Other possibilities arise, but I wanted to keep each graphic to 3-4 groups for analysis.
On the right, I calculated 1975-1990 averages for each class (restricting the average to somewhat isolate the factor difference from the trend.) As you see, there are substantial differences between the average temperature for each group. Autodin, NODC and buoy measurements run about 0.5 deg C warmer than ship measurements; sensor measurements (and there are a few different types) also run about 0.5 deg C warmer than buckets.
In the past, I’ve been sarcastic about the implausible assumption that all SST measurements changed from bucket to engine inlet in Dec 1941 after Pearl Harbor. (Hawaii doesn’t seem like a bad place to examine this assumption.) And indeed the form of the adjustment proposed by the climate science specialists is implausible. The transition from bucket to engine inlet began in the mid-1930s in this gridcell anyway and was merely completed during WW2.
ERSST contains an adjustment for this changeover – the need to an adjustment is evident and an appropriate adjustment isn’t all that easy to calculate given that there is a concurrent trend in temperature. The timing of some of the features in the ERSST record (relative to the averages of ICOADS data shown here) seem a bit odd.
The point that intrigues me though is the size of the difference between measurement types. So when one sees the warm recent values (in the period past the 1990 date to which I;ve collated ICOADS info), one wonders the degree to which the increase is due to inhomogeneous measurement methods as opposed to climatic change. (That the temperature has increased is not at issue – all that I’m observing here is that it is non-trivial allocating the increase in the sample average between changes in measurement type and climatic change.
Under such circumstances, it would be nice to consult the land record on the island to see if it can assist. Here’s a comparison of ERSST3 to CRUTEM for this gridcell (also compared on an earlier occasion). As readers are well aware, Phil Jones and CRU have supposedly “proven” one more time that there isn’t any material UHI impact on CRU records, though the non-existence of UHI is not totally evident in this graphic.
Maybe, like Mann’s birstlecones, Honolulu airport is a “sweet spot” for detecting climate change 🙂
I’ll re-visit this topic in a few months, but this is it for now.