Friday, March 7, 2014

New Berkeley Earth temperature dataset vs existing datasets

The Berkeley Earth team released a new temperature analysis that includes both land and ocean surface temperatures.  They used their existing land data and merged it with HadSST data (note: not HadSST3 as I originally wrote), using kriging to interpolate temperatures where data did not directly exist.  In this, their methodology is similar to the recent Cowtan and Way (2013) paper, however, Cowtan and Way used HadSST3 for their ocean data.  I've compared their new results over the past 30 years (Jan 1984-Dec 2013) to GISS, HadCRUT4, NCDC, UAH, and Cowtan and Way's results, first standardizing all temperature anomalies to the 1981-2010 baseline.  Why the past 30 years?  Thirty years is generally considered the standard time period for measuring climate.  All trends mentioned in this article are calculated using linear regression corrected for autocorrelation.


BEST vs. GISS


Since 1984, the new BEST temperature data shows significant temperature rise of 0.1776 ± 0.0469ºC per decade (trend ± 95% confidence interval).  That is faster than the rate of rise in GISS, which shows a significant rise of 0.1696 ± 0.0468ºC per decade.  However, given how the 95% confidence intervals overlap, the difference is not significant.

BEST vs. UAH


The rate of temperature rise since 1984 in UAH data  is very similar to that in BEST data.  Only 0.0010ºC per decade separates the two.  BEST: 0.1776 ± 0.0469ºC per decade, UAH: 0.1786 ± 0.0732ºC per decade.  Again, the difference is not significant.

BEST vs. Cowtan and Way 2013 




The Cowtan-Way dataset shows faster warming than the BEST data over the past 30 years and the fastest warming of all six data sets.  CW: 0.1919 ± 0.0496ºC per decade, BEST: 0.1776 ± 0.0469ºC per decade.  Again, however, the difference in the rate is not significant.

BEST vs. HadCRUT4


The warming rate for HadCRUT4 over the past 30 years clocked in at 0.1666 ± 0.0529ºC per decade, over 0.01ºC per decade slower than the BEST rate.

BEST vs. NCDC

 
Over the past 30 years, NCDC shows the least warming of the six data sets: 0.1612 ± 0.0470ºC per decade, a full 0.0164ºC per decade lower than BEST.

For those who want a nice summary of the differences in the warming rates, this table summarizes everything nicely:

Data setWarming since 1984 (± 95% confidence interval)
Cowtan-Way0.1919 ± 0.0496ºC per decade
UAH0.1786 ± 0.0732ºC per decade
BEST0.1776 ± 0.0469ºC per decade
GISS0.1696 ± 0.0468ºC per decade
HadCRUT40.1666 ± 0.0529ºC per decade
NCDC0.1612 ± 0.0470ºC per decade

While there's a 0.0307ºC per decade difference between the fastest and slowest rates, that difference is not significant given overlap between the 95% confidence intervals.  Larger differences appear in shorter time frames, such as since 1997.

Data setWarming since 1997 (± 95% confidence interval)
UAH0.12131 ± 0.17334ºC per decade
Cowtan-Way0.11833 ± 0.10133ºC per decade
BEST0.09408 ± 0.08397ºC per decade
GISS0.08660 ± 0.08069ºC per decade
HadCRUT40.06258 ± 0.09641ºC per decade
NCDC0.05486 ± 0.06985ºC per decade

Although the differences in rates since 1997 are again not statistically significant, there is a pattern to the data, both in the rates since 1984 and the rates since 1997.  HadCRUT4 and NCDC are the only two which do not fill in the gaps between weather stations and are therefore missing much of the polar regions, which are the fastest warming regions on the planet.  Not too surprising, then, that those two data sets show the least warming.  GISS and NCDC share a flaw in how they process sea surface temperatures, with a failure to calibrate sea surface temperatures taken by ships versus buoys.  As Cowtan and Way noted, that calibration failure introduces a false cooling bias into the data since 2000.  UAH, Cowtan-Way, and BEST, the three data sets that correct both coverage gaps and the sea surface temperature calibration, all show the most warming.

The other point to note?  There is no pause in warming over the past 17 years.  All three surface data sets that correct for coverage gaps (Cowtan-Way, BEST, and GISS) show statistically significant warming since 1997.  The supposed pause is largely an artifact of coverage gaps in HadCRUT4 and a combination of temporary cooling from ENSO, volcanic aerosols, and a drop in solar output.

In summary, the new BEST data set largely confirms what Cowtan and Way noted in their paper—that coverage gaps in HadCRUT4 lead to underestimated warming rates, especially over the past 17 years during the supposed "pause".  Removing those gaps shows that global warming may have slowed slightly but there is no "pause". 

No comments:

Post a Comment