A lot has been made of a paper (Lyman et al, 2006) that appeared last year that claimed that the oceans had, contrary to expectation, cooled over the period 2003-2005. At the time, we (correctly) pointed out that this result was going to be hard to reconcile with continued increases in sea level rise (driven in large part by thermal expansion effects), and that there may still be issues with way that the new ARGO floats were being incorporated into the ocean measurement network. Now it seems as if there is a problem in the data and in the latest analysis, the cooling has disappeared.
Ocean heat content changes are potentially a great way to evaluate climate model results that suggest that the planet is currently significantly out of equilibrium (i.e. it is absorbing more energy than it is emitting). However, the ocean is a very big place and the historical measurement networks are plagued with sampling issues in space and time. Large scale, long term compilations globally (such as by Levitus et al, 2001; Willis et al, 2004) and regionally (i.e. North Atlantic) have indicated that the oceans have warmed in recent decades at pretty much the rate the models expected.
Since 2000, though, ARGO – which is a network of floats that move up and down in the ocean and follow the currents – has offered the potential to dramatically increase the sampling density in the ocean and provide, pretty much for the first time, continuous, well spaced data from the least visited, but important parts of the world (such as the Southern Oceans). Data on ocean heat content from these floats had been therefore eagerly anticipated.
Initial ARGO measurements were incorporated into the Willis et al, 2004 analysis, but as the ARGO data started to dominate the data sources from around 2003, Lyman et al reported that the ocean seemed to be cooling. These were only short term changes, and while few would confuse one or two anomalous years with a long term trend, they were a little surprising, even if they didn’t change the long term picture very much.
The news this week though is that all of that ‘cooling’ was actually due to combination of a faulty pressure reading on a subset of the floats and a switch between differently-biased observing systems (Update: slight change in wording to better reflect the paper). The pressure error meant that the temperatures were being associated with a point higher in the ocean column than they should have been, and this (given that the ocean cools with depth) introduced a spurious cooling trend when compared to earlier data. This error may be fixable in some cases, but for the time being the suspect data has simply been removed from the analysis. The new results don’t show any cooling at all.
Are we done then? Unfortunately no. Because of the paucity of measurements, assessments of ocean heat content need to use a wide variety of sensors, each with their own quirks and problems. Combined with switches in data sources over the years, there is a significant potential for non-climatic trends to creep in. In particular, the eXpendable BathyThermographs (XBTs – sensors that are essentially just thrown off the side of the ship) have a known problem in that they didn’t fall as quickly as they were originally assumed to. This gives a warm bias (see this summary from Ingleby and Palmer or the paper by Gouretski and Koltermann) , particularly in data from the 1970s before corrections were fully implemented. We are still going to have to wait for the ‘definitive’ ocean heat content numbers, however, it is important to note that all analyses give long term increases in ocean heat content – particularly in the 1990s – whether they include the good ARGO data or exclude the XBTs or not).
There are a number of wider lessons here:
- New papers need to stand the test of time before they are uncritically accepted.
- The ARGO float data are available in near real-time, and while that is very useful, any such data stream is always preliminary.
- The actual problem with these data was completely unknowable when Lyman et al wrote their paper. This is in fact very common given the number of steps required to create global data sets. Whether it’s an adjustment of the orbit of a satellite, a mis-calibration of a sensor, an unrecorded shift in station location, a corruption of the data logger or a human error, these problems often only get fixed after a lot of work.
- Anomalous results are often the driver of fundamental shifts in scientific thinking. However, most anomalous results end up being resolved much more straightforwardly (as in the case, or the MSU satellite issue a couple of years back).
Scientists working in a field build up a certain intuition about how things ‘work’. This intuition can come from a gut instinct, deep theoretical understanding, robust model results, long experience with observations etc. New results that fall outside of that framework often have a tough time getting accepted, but if they are solid and get subsequent support they will generally be incorporated. But that intuition is also very good at detecting results that just don’t fit. When that happens, scientists spend a lot of time thinking about what might be wrong – with the data, the analysis, the model or the interpretation. It generally pays to withhold judgment until that process is finished.