Comments on “Atmospheric controls on northeast Pacific temperature variability and change, 1900–2012” by Johnstone and Mantua
Synopsis: A preliminary examination of this paper suggests that the results – in particular the conclusion that rising greenhouse gases played little role in the warming of the US Pacific coastal states – rested very strongly on the authors’ selection of a single dataset and time period.
Johnstone and Mantua (J&M) studied the connection between climate variability in the northeastern Pacific and surface air temperatures, and reached the startling conclusion that “natural internally generated changes in atmospheric circulation were the primary cause of coastal NE Pacific warming from 1900 to 2012.” They base this conclusion on year-to-year associations between sea level pressure over the North Pacific, and surface air temperature over Washington, Oregon, and California. While acknowledging that anthropogenic greenhouse gases are important for global climate, they assert that nearly all of the warming in these states, and much of the surface ocean temperatures off the coast can be explained by natural changes in circulation over the North Pacific before 1940.
These results appear at first glance to contradict our recent paper which concluded that, accounting for natural variability, rising greenhouse gases could explain most of the warming in the Northwest (Washington, Oregon , and Idaho). We expect that we and other scientists will need some time to unravel this apparent contradiction, but as a first step, we note some important facts that may have a bearing on the conclusions J & M reached.
1) Importance of testing the selection of dataset. The conclusions of J & M reside heavily on the influence of sea level pressures (SLP) over the NE Pacific. There is not a definitive long-term dataset of SLP to base climate studies on. The figure at the right shows a comparison of five datasets of SLP, taken from the supplementary material in their paper. The datasets appear to agree well after about 1950, but differ greatly during the first few decades of the 20th century. From the visual inspection alone it is clear that the datasets do not agree on the sign of the linear trend between 1900 and 1940. Even since 1950, while the datasets track year-to-year changes in SLP, they differ substantially in terms of their linear trend (-0.93mb/100yr for NCAR, +0.078 mb/100yr FOR HadSLP2, and -0.3 mb/100 yr for 20CR for 1950-2012). Their dataset of choice, the red “NCAR”, appears to us to cause the opposite result from what one would get by choosing any of the other datasets, most notably that it is the only data that exhibits a significant long- term trend over the 1900-2012 period. Given the substantial differences in long-term trends of these different datasets, we expect that J & M would have reached dramatically different conclusions if they had used alternative data sources. As the long-term behavior of NCAR differs from the other data sources, this raises doubts about their conclusions that most of the warming is due to the phenomena they describe. The authors’ rationale for choosing the NCAR SLP dataset is that prior to 1940, it most closely maintains the correlation apparent between SLP and SST seen post-1940. Other SLP datasets, on the other hand, show a divergence between SLP and SST pre-1940. As the authors point out, this may be a useful indicator of problems with the non-NCAR datasets pre-1940, but it means that they’ve chosen the dataset with the tightest link between SLP and SST, and it is this link that forms the basis of their conclusions. This leads us to point #2…
2) Similar to #1, As J&M themselves note, their results are strongly influenced by what happened before 1940. Since most of the anthropogenic warming occurred after 1960, repeating the analysis for, say, 1940-2012 would test whether their result was a product of the decision to use the NCAR dataset or is a robust physical feature of the climate system. Unfortunately, their paper does not give an indication that they conducted such analysis. Multiple compilations of surface air temperature covering the Northwestern US document the observed warming over the past century, and that most of the warming has occurred over the last half-century, rather than prior to 1940.
3) Our study performed multiple linear regression, an approach to identifying (consistently and simultaneously) the relative role of a variety of factors that included the El Nino-Southern Oscillation, Pacific North American Pattern, volcanic aerosols, solar variability and anthropogenic forcing. (MLR has its faults, but at least it treats predictors similarly.) Their study, by contrast, used single linear regression to identify a relationship between SLP and air temperature, and then diagnosed the anthropogenic component as the linear trend in the residual (which was small) with anthropogenic forcing. The problems with this approach are (a) that it does not treat the two sources of variability consistently, and (b) the time dimension is reduced to a single degree of freedom, the slope. As the figure below shows, multiple linear regression assigns a large role to Pacific climate variability (ENSO, in this example) but the radiative forcing is also large, but plays a role only after 1960. We also examined these relationships by season, as the influence of natural climate variability on temperatures over the Northwestern US is most prominent during the cool season.
4) The authors fault the CMIP5 global models for failing to reproduce the observed trends in SLP – but the ‘observations’ themselves have opposite trends as noted in our point 1 and the first figure, even since 1950.
- John Abatzoglou, U Idaho email@example.com
- Philip Mote, OSU firstname.lastname@example.org
- David Rupp, OSU email@example.com