Changes in ocean heat content are thought to be the best way to detect and measure a planetary energy imbalance, since the vast majority (about 93%) of the extra heat goes into the ocean, and because its huge heat capacity -- about 1,000 times that of the atmosphere -- means heat changes there are much less fickle than in the atmosphere. As oceanographer Greg Johnson of NOAA puts it, "global warming is ocean warming."
OHC for the top half of the ocean (0-2000 meters), measured by the Argo bouys in the last 10+ years, is now clearly accelerating. The year-over-year change for the 0-700 meter region is 1.1 W/m2, and 1.5 W/m2 for the 0-2000 meter region.
For the 0-2000 meter region, a quadratic fit to the data is better than a linear fit, with an acceleration of 0.09 ± 0.03 (W/m2)/yr (statistical error, no autocorrelation):
Here, ZJ = 1 zettajoule = 1021 Joules.
A quadratic fit to these data keeps getting better and better relative to a linear fit:
4 comments:
Well, let's wait for the article. ;-)
Not so sure if the improvement in the fit justifies the more complicated quadratic statistical model. And then you should still study whether trends caused by measurement errors are sufficiently small; just statistics is not enough.
Part of the apparent acceleration is probably caused by geographical coverage deficiencies in the Argo network producing variable biases depending on spatial warming patterns. Similar to how Cowtan & Way diverges strongly from HadCRUT4 between 2005 and 2012, but the divergence has closed in the past couple of years.
As noted by von Schuckmann et al. 2014 the coverage gap around Indonesia means an area of strong ocean warming between 2005 and 2012 was not recorded. Thereafter variability has flipped and there has been stronger warming elsewhere.
Dramatic sea ice losses between 2005 and 2012 indicate a sustained increased heat flux into the Arctic, which has since subsided. The Arctic is also not covered by Argo.
Agree with Victor. That's a very marginal increase in R^2 but with one more parameter, and with n ~ 45 or so. I doubt that AIC or BIC analysis would return the 2nd order poly as the best model. You could try eliminating the x^1 term from it and then it might.
Jim: How does one do a regression analysis while forcing the linear term (x^1) to be zero?
Post a Comment