Do the Adjustments to Sea Surface Temperature Data Lower the Global Warming Rate?

Quick answer:  Over the long term, the answer is yes, but for shorter terms it depends on the sea surface temperature dataset and time period.  And in recent years, as most people understand, the adjustments increase global warming trends.

INTRODUCTION

When discussing global surface warming, we often read comments around the blogosphere to the effect of We need to examine the “raw” sea surface temperature data before all of the adjustments.  Typically, someone else will reply, The “raw” sea surface temperature data have a higher warming rate than the “adjusted” data, or written another way, The adjustments lower the sea surface temperature trends.   While that is true for the long term, it is not necessarily true for shorter time periods.  It depends on the dataset and timeframe.

This is the first in a series of posts about adjustments to global surface temperature data.  In the second, we’ll present the impacts of the adjustments to land surface air temperature data.

DATA

For the following illustrations, we’ll compare the linear trends of the “raw” sea surface temperature data to the warming rates of the “adjusted” datasets that are or had been used in global land+ocean surface temperature products and that are still being updated.

ICOADS – This is the source sea surface temperature data for the NOAA and UKMO sea surface temperature products.  It will serve as the “raw” data.

The others are the “adjusted” data.  The adjustments are said to be corrections to account for biases attributed to different types of temperature sampling methods:

• wooden, canvas and insulated buckets tossed overboard and hauled back on deck where thermometers are placed within the buckets,
• engine room inlets from differing depths
• moored and drifting buoys
• satellite measurements (limited to HADISST and Reynolds OI.v2 starting in the early 1980s)

Another type of adjustment is infilling…assigning temperature values to ocean grids without source data.  That is, during any given month, an ocean grid may not include observations-based data.  That lack of data grows worse as we travel back in time.  If an ocean grid does not contain data, the data supplier can leave that grid without data (as the UKMO does with its HADSST3 data), or the supplier can fill in that grid with make-believe data that is based on a variety of statistical methods. The UKMO infills its HADISST data and NOAA infills its ERSST.v3b, ERSST.v4 and Reynolds OI.v2 data…though the Reynolds OI.v2 data also include satellite-based observations, which drastically reduces the amount of infilling. However, for the not-infilled HADSST3 data from the UKMO, when the monthly data for a hemisphere are later averaged by UKMO (it’s a weighted average that considers latitude), the average is indirectly assigned to the grids without data.

The “adjusted” data presented in this post are:

ERSST.v4 – This is NOAA’s “pause buster” sea surface temperature dataset.  NOAA/NCEI and NASA/GISS have been using it in the global land+ocean surface temperature products since mid-2015.  The NOAA ERSST.v4 data have been adjusted for ship-buoy biases.  Those ship-buoy bias adjustments have their greatest impacts after the early 2000s. We discussed the numerous oddities in the ERSST.v4 data in the posts here and here.  The ERSST.v4 data are infilled and rely on in situ data only (no satellite-based data).

HADISST/Reynolds – GISS used to merge two sea surface temperatures datasets (UKMO’s HADISST and NOAA’s Reynolds OI.v2) for their Land-Ocean Temperature Index (LOTI). (GISS switched to NOAA’s ERSST.v3b data for their “official” sea surface temperature dataset in early 2013 and then switched again to NOAA’s ERSST.v4 data in mid-2015.) Basically, for the merged dataset, the HADISST data were used from 1880 to 1981 and Reynolds OI.v2 from 1982 to present.  See the GISS Surface Temperature Analysis webpage under the heading of Step 4 : Reformat sea surface temperature anomalies.  The merged HADISST/Reynolds sea surface temperature data are still an option for the GISS LOTI data.

Note 1: The NOAA Reynolds OI.v2 (and UKMO HADISST) data also use satellite-derived data from the early 1980s to present. Of the datasets presented in this post, the Reynolds OI.v2 portion of the HADISST/Reynolds data is the only one to use satellite-based data.

Note 2:  To confuse matters more for those new to sea surface temperature data, NOAA produces two versions of the satellite-enhanced Reynolds OI.v2 sea surface temperature data.  The original, used by GISS until early 2013, is prepared monthly and weekly with a 1-degree spatial resolution. This original version was not corrected for ship-buoy biases.

The other version of the Reynolds OI.v2 data is the high-resolution (1/4-degree) daily version, which is preferred by alarmists. It has been corrected for ship-buoy biases, but NOAA’s corrections are so heavy-handed that the warming rate of the high-resolution daily Reynolds OI.v2 data during a NOAA-selected “hiatus” period is far above the parametric uncertainty range of NOAA’s ERSST.v4 “pause-buster” data.  See the discussion of Figure 6 from the post On the Monumental Differences in Warming Rates between Global Sea Surface Temperature Datasets during the NOAA-Picked Global-Warming Hiatus Period of 2000 to 2014.

[End notes.]

HADSST3 – This is the sea surface temperature dataset used by the UK Met Office for their HadCRUT4 land+ocean surface temperature dataset.  Like the NOAA ERSST.v4 data, HADSST3 has been corrected for ship-buoys biases.  And as noted earlier, the ship-buoy bias adjustments have their greatest impacts after the early 2000s.  Where the NOAA ERSST.v4 data (and the HADISST, Reynolds OI.v2 and ERSST.v3b data) are infilled, the HADSST3 data are not.  HADSST3 data also exclude satellite-based data.

As an afterthought, I’ve listed the trends of NOAA’s ERSST.v3b data on the graphs but did not include their curves.  The ERSST.v3b was the predecessor to NOAA’s pause buster data.  It was used by NOAA for their combined land+ocean surface temperature data from 2009 to mid-2015.  ERSST.v3b was also used by GISS starting in early 2013 and ending in mid-2015.   Curiously, NOAA is still updating ERSST.v3b even though it was replaced by the ERSST.v4 “pause-buster” data, and, as a result, ERSST.v3b is still available as an option for the GISS LOTI dataset. The ERSST.v3b data are infilled and in situ only (no satellite-based data).

The base years for anomalies for all of the graphs are 1981-2010.  The top graphs include the trend lines.  I’ve excluded them in the bottom graph for those who want a clearer view of the curves of the individual datasets.

The “global” oceans are limited to the latitudes of 60S-60N, because the suppliers account for sea ice differently. Excluding the polar oceans is commonly done in scientific studies that compare sea surface temperature data.

The source of data used in the post is the KNMI Climate Explorer.

LONG-TERM COMPARISON

Figure 1 includes the “raw” and “adjusted” global sea surface temperature data for the period of 1880 to 2015.  (The sea surface temperature data extend farther back in time, but I’ve used 1880 as the start year, because the global land+ocean surface temperature products from NOAA and GISS start then.)

Figure 1

As shown, the “raw” sea surface temperature data have a higher warming rate than the “adjusted” data, because the “raw” data are biased cool prior to about 1940.  That “cool” bias is said to result from changes in methods used to sample ocean surface temperatures before World War II.

Note: The excessive warming of the “raw” data from the early 1900s to the mid-1940s presented a problem for climate models. Most of the warming then was not caused by man-made greenhouse gases (according to the models), but the warming trend of the “raw” data from the early 1900s to the mid-1940s was much higher than their recent warming rates.  For confirmation, see the graph of 30-year running trends here.  The bias corrections for the data prior to 1940 reduced those problems for the models, but did not eliminate them.  That is, the models still cannot explain the initial cooling of global sea surface temperatures from 1880 to about 1910, and, as a result, the models cannot explain the warming from about 1910 to the mid-1940s. And you’ll note for all of the “adjusted” sea surface temperature datasets, the 30-year trends ending about 1945 are higher than the most recent 30-year trends.   [End note.]

The UKMO HADSST3 and the two ERSST from NOAA show the same warming rates from 1880 to 2015.  The outlier of the adjusted data for the long-term is HADISST/Reynolds.  The HADISST portion shows less initial cooling from 1880 to about 1910.

TRENDS FROM 1950 TO 2015

For the first of the shorter-term periods, we’ll use one of the start years used by NOAA in Karl et al (2015) Possible artifacts of data biases in the recent global surface warming hiatus.  That start year is 1950.  (See Figure 1 from Karl et al (2015).)  And we’ll end the data with the most-recent full year, 2015.

For the period 1950 to 2015, Figure 2, the two NOAA ERSST reconstructions (ERSST.v3b and ERSST.v4 “pause buster”) have higher warming rates than the “raw” data from ICOADS. In other words, for those two datasets, the adjustments increased the warming rates versus the “raw” data.  Why?  Neither of those two datasets from NOAA have been adjusted for the 1945 discontinuity (and trailing biases) presented in Thompson et al. (2008)  A large discontinuity in the mid-twentieth century in observed global-mean surface temperature.  For a more-detailed discussion of NOAA’s failure to account for those biases with their ERSST.v4 “pause-buster” data, see the post Busting (or not) the mid-20th century global-warming hiatus, which was also cross posted at Judith Curry’s ClimateEtc here and at WattsUpWithThat here.

Figure 2

On the other hand, the HADSST3 data from the UKMO have been corrected for the 1945 discontinuity and trailing biases presented in Thompson et al. (2008), and as a result, the HADSST3 data have a lower warming rate than the “raw” data from 1950 to 2015. Curiously, the merged HADISST/Reynolds OI.v2 data also have not been adjusted for the 1945 discontinuity, yet they too have a lower warming rate than the “raw” source data.

Note: For those interested in a comparison for the period of 1945 (start of the discontinuity period) to 1975 (commonly used as the breakpoint between recent warming period and mid-20^th Century slowdown) see the supplemental graph here. [End note.]

TRENDS FROM 1975 TO 2015

As noted immediately above, the year 1975 is commonly used as a breakpoint between recent warming period and mid-20^th Century slowdown in global surface warming.

Figure 3 presents the trends from the 1975 start of the recent warming period to 2015.  Only the UKMO’s HADSST3 data have a higher warming rate than the “raw” data during this period.  The warming rates of the “adjusted” data during this period range from about 0.10 deg C/decade to about 0.14 deg C/decade.

Figure 3

Note:  1979 is an often-used start year for surface temperature data, especially when compared to lower troposphere temperature data, which begin then.  If we were to start Figure 3 in 1979, we’d get similar results, inasmuch as HADSST3 is the only dataset with a warming rate that’s higher than the “raw” data. [End note.]

TRENDS FROM 1998 TO 2015

1998 is often used as the start year in presentations of the slowdown in global warming. 1998 was also used by Karl et al. (2015) as a start year for trend comparisons. (Again refer to Figure 1 from Karl et al. (2015).)  During the period of 1998 to 2015, with the exception of the NOAA ERSST.v3b data (curve not shown), all “adjusted” sea surface temperature datasets show higher warming rates than the “raw” data. See Figure 4.  One curiosity is that the warming rate of the HADSST3 data (which have been adjusted for ship-buoy biases) is comparable to the trend of the satellite-enhanced Reynolds OI.v2 portion of the HADISST/Reynolds data (which have not been adjusted for ship-buoy biases).

Figure 4

But the real oddity exists when we compare the trends of the two datasets that have been corrected for ship-buoy biases and they are the UKMO HADSST3 and NOAA’s “pause-buster” ERSST.v4. Recall that the ship-buoy bias is said to have its greatest impact after the early 2000s. For the HADSST3 data, that correction only added about +0.01 deg C/decade to the warming rate from 1998 to 2015, but for NOAA’s ERSST.v4 “pause buster” data, the ship-buoy bias corrections added about +0.05 deg C/decade to the warming rate.  That additional warming in the ERSST.v4 data, of course, was caused by NOAA tweaking all of the tuning knobs (parameters) in their sea surface temperature model so that the hiatus warming rate was near to the high end of the parametric uncertainty range.  (See the post The Oddities in NOAA’s New “Pause-Buster” Sea Surface Temperature Product – An Overview of Past Posts and the posts linked therein for further information.)

As seems to be the fad recently, I’m sure that someone’s going to compare the trends for the period of 1975 to 2015 (Figure 3) to those of the period of 1998 to 2015 (Figure 4) and say a slowdown in global warming didn’t exist. As with everything else, that depends on the sea surface temperature dataset.  One problem with that comparison is that any slowdown from 1998 to 2015 is included in the data from 1975 to 2015.  Another problem is that the period of 1998 to 2015 includes naturally caused upticks in 2014 and 2015. So as a reminder…

WAS THERE A HIATUS OR SLOWDOWN IN GLOBAL WARMING IN SEA SURFACE TEMPERATURE DATA?

Of course there was, but the extent of the slowdown depends on the sea surface temperature dataset and the period to which the slowdown is compared.  For example, Figure 5 includes the “raw” and “adjusted” global sea surface temperature anomalies for the period of 1998 to 2013.  We ended the data in 2013, because:

• 2013 was an ENSO neutral year…that is there no El Niño or La Niña. (See NOAA’s Oceanic NINO Index here.)
• The Blob and the weak El Niño conditions were the primary causes of the naturally occurring uptick in global sea surface temperatures in 2014 and,
• The continuation of The Blob and the strong El Niño conditions were the primary causes of the naturally occurring uptick in global sea surface temperatures in 2015.

Figure 5

Note 1: To confirm the second and third bullet points, we discussed and illustrated the natural causes of the 2014 “record high” surface temperatures in General Discussion 2 of my free ebook On Global Warming and the Illusion of Control (25 MB).  And we discussed the naturally caused reasons for the record highs in 2015 in General Discussion 3.

Note 2: Some may claim the start year of 1998 is cherry-picked because it’s an El Niño decay year. That’s easily countered by noting that the 1997/98 El Niño was followed by the 1998 to 2001 La Niña. (Once again, see NOAA’s Oceanic NINO Index here.) Also, 1998 was used as a start year by Karl et al. (2015).

[End notes.]

Karl et al. (2015) also used a sleight of hand in their trend comparisons by using 1950 as the start year of the recent warming period. The IPCC did the same thing in their analyses of it in Chapter 9 of their Fifth Assessment Report (See their Box 9.2).  Both groups referenced the hiatus warming rates to the warming rate of periods starting in 1950 or 1951.  Why does that indicate they were using smoke and mirrors?  The trends from those periods with 1950 or 1951 start dates include the slowdown or cooling of ocean surfaces that occurred from the mid-1940s to about 1975. (See the supplemental graph here for the trends from 1950 to 1975).  The only datasets where it makes little difference are the NOAA ERSST data; then again, that’s caused by NOAA failing to correct for the 1945 discontinuity and trailing biases.  So, let’s present the trends from the start of the recent warming period (1975) to the end of the 20^th Century (1999) solely as an example. See Figure 6.

Note: I am not suggesting in any way that 1999 is the breakpoint between the late-20th Century warming period and the slowdown.  I am simply using the year 1999 because it was used by NOAA in Karl et al. (2015). (Again refer to Figure 1 from Karl et al. (2015).) [End note.]

Figure 6

Only the UKMO’s HADSST3 data have a higher warming rate than the “raw” data during this period.

Now let’s compare the trends for the periods of 1975 to 1999 (Figure 6) and 1998 to 2013 (Figure 5).  Of the adjusted datasets, the greatest slowdown occurs with the HADSST3 data, with the trend for 1998 to 2013 being almost 0.14 deg C/decade lower than the 0.16 deg C/decade trend for the period of 1975 to 1999. (Part of that decline in the HADSST3 data was caused by its relatively high warming rate from 1975 to 1999.) The obsolete ERSST.v3b data is next with a drop in the warming rates between the two periods of roughly 0.12 deg C/decade.  The HADISST/Reynolds data showed the least warming for 1975 to 1999 so its decline in trend is about 0.09 deg C/decade.  Then there’s the NOAA ERSST.v4 “pause-buster” data, which shows the smallest slowdown, with the trend for 1998 to 2013 only being 0.06 deg C/decade lower than the 0.13 deg C/decade trend for the period of 1975 to 1999.

The “raw” data show a slowdown of about 0.14 deg C, which is the same as the slowdown in the HADSST3 data.

CLOSING

The title question of the post was Do the Adjustments to Sea Surface Temperature Data Lower the Global Warming Rate?

For the period of 1880 to 2015 (or for any period ending in recent times that starts well before the 1940s) the answer is yes.  As shown in Figure 1, all “adjusted” sea surface temperature datasets have lower warming rates than the “raw” data.

For the period of 1950 to 2015, Figure 2, only the NOAA ERSST datasets have higher warming rates than the “raw” source data, because they were not adjusted for the 1945 discontinuity and trailing biases presented in Thompson et al. (2008).  The NOAA ERSST.v3b data were published about the same time as Thompson et al. so we would not have expected NOAA to include those adjustments. On the other hand, the NOAA ERSST.v4 data were published well after Thompson et al., but NOAA failed to account for the biases documented in that paper.

For the period of 1975 to 2015, Figure 3, only the UKMO HADSST3 data have a higher warming rate than the “raw” source data from ICOADS. Why?  Dunno.

Then for the period of 1998 to 2015, with the exception of the ERSST.v3b data (not the “pause-buster” ERSST.v4 data), the warming rates of all “adjusted” sea surface temperature datasets exceed the trend of “raw” source data.

So the next time someone claims, without defining the timeframe, that the adjustments to sea surface temperature decreases global warming, you can answer:

Over the long term, that’s correct, but for shorter terms it depends on the sea surface temperature dataset and time period. And for the period after 1998, it’s wrong.  The adjustments definitely increase the global warming rate, with the “pause-buster” ERSST.v4 sea surface temperature data having the greatest increase.

Next, we’ll look at the land surface air temperature data.