by Zeke Hausfather and Kevin Cowtan
A buoy-only sea surface temperature record supports NOAA’s adjustments.
Significant recent media and political attention has been focused on the new NOAA temperature record, which shows considerably more warming than their prior record during the period from 1998 to present. The main factor behind these changes is the correction in ocean temperatures to account for the transition from ship engine room intake measurement to buoy-based measurements and a calibration of differences across ships using nighttime marine air temperatures (NMAT). Here we seek to evaluate the changes to the NOAA ocean temperature record by constructing a new buoy-only sea surface temperature record. We find that a record using only buoys (and requiring no adjustments) is effectively identical in trend to the new NOAA record and significantly higher than the old one.
The changes to the prior NOAA global land/ocean temperature series are shown in Figure 1. There are some large changes in the 1930s that are interesting but have little impact on century-scale trends. The new NOAA record also increases temperatures in recent years, resulting a in a record where the period subsequent to 1998 has a trend identical to the period from 1950-1997 (and giving rise to the common claim that the paper was “busting” the recent slowdown in warming).
Figure 1: New and old homogenized global land/ocean records from Karl et al, 2015.
The paper that presented the revised record, Karl et al, didn’t actually do much that was new. Rather, they put together two previously published records: an update to the NOAA sea surface temperature record (called ERSST) from version 3 to version 4, and the incorporation of a new land record from the International Surface Temperature Initiative (ISTI) that makes use of around 32,000 land stations rather than the 7,000 or so GHCN-Monthly stations previously utilized. The new land record is quite similar to that produced by Berkeley Earth, though it has relatively little impact on the temperature trend vis-à-vis the old land record, particularly during the recent 1998-present period.
The slowdown-busting nature of the Karl et al paper relies almost entirely on the update from ERSST v3 to v4. During the post 1998 period, this is primarily due to a revised treatment of buoys and ship engine room intake (ERI) measurements and an improved calibration of differences across ships. During the past few decades the number of automated SST measurement buoys has expanded rapidly from effectively zero before 1980 to over 70 percent of all SST measurements today as shown in the figure below. Buoys are appealing measurement platforms, as they are not restricted to shipping routes and often have fully automated reporting via satellite uplink.
Figure 2: Share of SST observations by instrument type from Kennedy et al 2011. Note that this figure ends in 2006; since they buoys have continued to grow in observation share.
NOAA argues that the transition to buoys introduced a spurious cooling bias into the record. ERIs tend to warm the water a bit before measuring it (ship engine rooms being rather hot), whereas buoys do not. They identify a bias of around 0.1 C between buoys and ERIs and remove it by adjusting buoy records up to match ERI records in ERSST v4, as well as use NMAT readings to calibrate the differences across ships. These adjustments had not been done in the prior ERSST v3. As an aside, the decision to adjust buoys up to ERIs or ERIs down to buoys should nominally be trend neutral. Indeed, in their work on HadSST3 Kennedy and colleagues explicitly tested this, and found “no appreciable difference” on trends.
However, there is a rather straightforward way for us to test if the adjustments done in ERSSt v4 are proper or not: compare their adjusted record to a record made only from buoys. The buoy records are from purpose built instruments which are largely standardized, resulting in much more homogeneous records. On the other hand, the buoy record is short, and has limited coverage in the early 90’s.
The buoy-only record is prepared by calculating daily averages for each buoy. Buoys which show a large daily temperature variation are rejected: in deep water the daily temperature range is only a few tenths of a degree, but in very shallow water it can be substantial which presents problems when some data are missing. Next, the daily data are placed into 550 km equal area grid cells based on the location of the buoy for that day, and monthly averages are determined for each cell.
The resulting coverage is still limited and so produces a biased estimate of global sea surface temperature. To produce a useful comparison to ERSST, we therefore reduce the coverage of the ERSST datasets to match the buoy dataset (now using a fine 1 degree grid for all the data) and then calculate anomalies for all the datasets using a 2001-2010 baseline. The area weighted mean temperature is then calculated for each record. While this doesn’t provide a very good estimate of global SST, it does allow a strict like-with-like comparison against ERSST over the regions where they buoys have coverage. The percent of global ocean covered by buoy measurements varies from around 40% in the mid 1990s to around 70% in recent years.
Figure 3: ERSST v3, v4, and Buoy-Only SST anomalies and trends from 1995 through the end of 2014. The trend periods shown are the full record (1995-2014) and the “hiatus” period (1998-2014). 2015 is excluded as the year is incomplete, and the period prior to 1995 is excluded due to limited buoy coverage. The anomaly graph is baselined to 1995-2005 to show the time-evolution of differences.
As shown in Figure 3, a buoy-only record is quite similar to the ERSST v4 but shows much more warming than ERSST v3 during the period from 1995 to present. This suggests that ERSST v3 suffered a cooling bias when blending buoy and ship records that is properly corrected in ERSST v4, at least for the areas where both ship and buoy records are available. Because the buoy record is relatively homogenous and requires no adjustments, it provides a good check in the validity of the combined ship-buoy series when normalized for spatial coverage.
The ship records are important because they form the foundation for a long sea surface temperature record, but they require careful calibration. The differences between HadSST3 and ERSSTv4 suggest that the finer details of the ship record are not yet settled, and as a result care is required especially when considering short term trends. However the buoy data support the NOAA claim that ERSSTv3 suffered a significant cool bias over recent years arising from inhomogeneities in the ship record and the increasing use of buoys.
Code for downloading and processing the data for this analysis is available here. Note that the underlying buoy dataset is large (approximately 44 GB). Gridded 1×1 files are also provided at for buoy, ERSSTv3, and ERSSTv4 data.
Zeke and Kevin have taken a useful first step in deciphering what is going on with NOAA’s new ERSSTv4 data set.
The global ocean trend from 1998-2014 is cited in Karl et al. SOM as 0.097 C/decade for ERSSTv4 and 0.038 C/decade for ERSSTv3b. In Figure 3 above, the red dot for ERSSTv4 for the regions that are sampled by buoys is about 0.12 C/decade. Note also the slightly lower trend in Figure 3 for the buoys only (~0.102 C/decade).
There are substantial changes between v3 and v4 during this period, and the buoys certainly seem to contribute to a larger trend.
It is instructive to compare the change between HADSST2 and HADSST3, which also included a buoy adjustment in HADSST3:
The Huang et al. paper (from NOAA) does some comparisons of ERSSTv4 with HADSST3 for longterm trends and a few regions, but not for this particular periode. The relevant period is compared in this plot by Bob Tisdale, showing a substantially larger trend for ERSSTv4:
You may not be aware that NOAA has another SST dataset – the OISSTv2 – for the period since the 1980’s. This data set has higher horizontal resolution owing to the use of satellites, and references satellite and ship observations to the buoys. Serendipitously, I spotted this article by NOAA this morning on twitter: Exactly the same but different: why we have so many different ways of looking at sea surface temperature. It’s a good article, and describes the advantages of OISST v2. Criticisms of OISST are that the satellite data aren’t consistent across the period. Well, the satellites have been pretty consistent since 1995, and the point of the buoys is to calibrate the satellite observations to adjust for any satellite inconsistencies, volcanic eruptions, etc.
Below is a plot of the OISST v2 (provided by Bob Tisdale).
Bob Tisdale cites a trend in OIv2 of 0.041 C during the period 1998-2014.
The bottom line seems to be that the NOAA ERSSTv4 1998-2014 trend is about twice as large as the trend for HADSST3 and OISSTv2.
So, why are these 3 analyses, primarily based on buoys during this period, so different? A way to approach understanding this would be to do the same buoy masking on the HADSST3 and OIv2 data sets; this would help assess/understand the differences.
p.s. Given the controversial nature of the Karl et al. paper, and the fact that I had written previously on this topic, I thought it best to include my comments in the main post, rather than in the comments thread. Since this is a guest post, please keep your comments relevant and civil.