This post compares the instrument observations of three global temperature anomaly datasets (NINO3, Global, and North Atlantic “Plus”) to the hindcasts of the NCAR couple climate model CCSM4, which was used in a couple of recent peer-reviewed climate studies. Those studies relied solely on the models and do not present time-series graphs that compare observational data to the 20thCentury hindcasts to allow readers to determine if the NCAR CCSM4 model has any basis in reality. So far, I have not seen this done in any of the blog posts that have discussed those studies.
This post would have been much easier to prepare if the Sea Surface Temperature outputs of the NCAR CCSM4 were available through the Royal Netherlands Meteorological Institute (KNMI) Climate Explorer website. Then I would not have felt obligated to provide as many introductory explanations and graphs, and supplemental comparisons. (The post would have been easier to write, and easier to read.) But since the modeled Sea Surface Temperature data are not available yet, I’ll present, for the time being, the NCAR CCSM4 hindcast surface air temperature anomalies. As noted a number of times throughout this post, if and when the Sea Surface Temperature hindcasts are available through the KNMI Climate Explorer, I will be more than happy to update this post.
Note 1: The period used in this post runs from January 1900 to December 2005 because the NINO3 Sea Surface Temperature anomalies, used in one of the studies, have little to no source data before 1900, and the CCSM4 model hindcast ends in 2005.
Note 2: The source of data for this post, as noted above, is the KNMI Climate Explorer. Surface Air Temperature is available for the NCAR CCSM4 on their Monthly CMIP5 scenario runs webpage, but Sea Surface Temperature (identified as TOS) is not. Therefore, this post compares the Surface Air Temperatures anomalies (which over the oceans would be comparable to Marine Air Temperature anomalies) of the model outputs to Sea Surface temperature anomalies during the discussion of ENSO. And as you will see, this should not present any problems for this discussion. For the model-to-data comparisons of global and of North Atlantic “Plus” surface temperature anomalies, observed land plus sea surface temperature anomalies are compared to the modeled Surface Air Temperature anomalies for land and oceans. This is common practice in posts that compare instrument observations to model outputs at blogs such as Real Climate (Example post: 2010 updates to model-data comparisons) and Lucia’s The Blackboard, (Example post: GISTemp: Up during August!), and it assumes the modeled sea surface temperatures will be roughly the same as the modeled Marine Air Temperatures. (More on this at the end of the post.)
Note 3: This post does not examine the projections of future climate presented in the referenced papers. This post examines how well or poorly the CCSM4 ensemble members and model mean match the observations that are part of the instrument temperature record. You, the reader, will then have to decide whether the model-based studies that use the CCSM4 are of value or whether they should be dismissed as mainframe computer-crunched conjecture.
The National Center for Atmospheric Research (NCAR) Community Climate System Model Version 4 (CCSM4) coupled climate model has been submitted to the Coupled Model Intercomparison Project Phase 5 (CMIP5)archive of coupled climate model simulations. (Phew, try saying that fast, three times.) And much of the data provided to CMIP5 for those models are presently available through the KNMI Climate Explorer Monthly CMIP5 scenario runs webpage. The model data stored in the CMIP5 archive will serve as a source for the next IPCC report, AR5, due in 2013. Refer to the Real Climate post CMIP5 simulationsfor further information.
Two papers based on the NCAR CCSM4 climate model have recently been published. There may be other published papers as well based on the CCSM4. The first is Meehl et al (2011) “Model-based evidence of deep-ocean heat uptake during surface-temperature hiatus periods”, paywalled. Meehl et al (2011) is a model-based study that attempts to illustrate that ocean heat uptake can continue during decadal periods when global surface temperatures flatten. The La Niña portion of the natural climate phenomenon called the El Niño-Southern Oscillation (ENSO) was determined to be the possible cause for the flattening of surface temperatures and for the increase in Ocean Heat uptake. Yup, La Niña events. For me, the paper raised a number of questions. One of them was: Why did Meehl et al (2011) only discuss decadal hiatus periods, during which surface temperatures failed to rise? The instrument temperature record for 20thCentury clearly shows a multidecadal decline in surface temperatures from the mid-1940s to the mid-1970s that is attributable, in part, to a mode of natural variability called the Atlantic Multidecadal Oscillation. Doesn’t the NCAR CCSM4 simulate multidecadal variability?
(For an introductory discussion of the Atlantic Multidecadal Oscillation, refer to the post An Introduction To ENSO, AMO, and PDO — Part 2.)
The second paper is Stevenson et al (2011) “Will there be a significant change to El Niño in the 21st century?”, also paywalled, but a Preprint exists for it. Stevenson et al (2011) is also a model-based study that attempts to illustrate, from the abstract:
“ENSO variability weakens slightly with CO2; however, various significance tests reveal that changes are insignificant at all but the highest CO2 levels.”
In other words, there is little change in the model depiction of the strength and frequency of El Niño and La Niña events with increasing levels of anthropogenic greenhouse gases. The Stevenson et al (2011) abstract concludes with:
“An examination of atmospheric teleconnections, in contrast, shows that the remote influences of ENSO do respond rapidly to climate change in some regions, particularly during boreal winter. This suggests that changes to ENSO impacts may take place well before changes to oceanic tropical variability itself becomes significant.”
The NCAR “Staff Notes” webpage “El Niño and climate change in the coming century” provides this further explanation:
“However, the warmer and moister atmosphere of the future could make ENSO events more extreme. For example, the model predicts the blocking high pressure south of Alaska that often occurs during La Niña winters to strengthen under future atmospheric conditions, meaning that intrusions of Arctic air into North America typical of La Niña winters could be stronger in the future.”
I suspect we’ll be reading something to the effect of “oh, the cold temperatures were predicted by climate models”, referring to Stevenson et al (2011), if the coming 2011/12 La Niña winter is colder than normal in North America.
Since the El Niño-Southern Oscillation (ENSO) is a major part of both papers, let’s start with it. For those new to ENSO, refer to the post “An Introduction To ENSO, AMO, and PDO – Part 1”for further information.
HOW WELL DOES CCSM4 HINDCAST CERTAIN ASPECTS OF ENSO?
Stevenson et al (2011) used NINO3 Sea Surface Temperature (SST) Anomalies as their primary El Niño-Southern Oscillation index. NINO3 is a region in the eastern equatorial Pacific with the coordinates of 5S-5N, 150W-90W. Its sea surface temperature anomalies are used as one of the indices that indicate the frequency and magnitude of El Niño and La Niña events. Unfortunately, the CCSM4 modeled Sea Surface Temperature data are not available for download through the KNMI Climate Explorer as of this writing. That requires us to use the model’s Surface Air Temperature output in the comparisons. The second problem is that the comparable instrument observations dataset, Marine Air Temperature, for the NINO3 region becomes nonexistent before 1950, as shown in Figure 1, and we’d like the comparison to start earlier than 1950.
The other option is to compare Sea Surface Temperature observations for the NINO3 region to the Surface Air Temperature outputs of the model. This should be acceptable in the equatorial Pacific since there is little difference between the observed NINO3 Sea Surface Temperature anomaly and Marine Air Temperature anomaly data. Figure 2 compares observed Sea Surface Temperature anomalies and Marine Air Temperature anomalies from January 1950 to December 2005. As illustrated, there are differences in the magnitude of the year-to-year variability between the two datasets. The Sea Surface Temperature anomalies vary slightly more than the Marine Air Temperature anomalies. But the timing of the variations are similar, as one would expect. The correlation coefficient for the two datasets is 0.92. Also note that the linear trends for the two datasets are basically identical. As mentioned earlier, the comparison of NINO3 Sea Surface Temperature Anomaly observations to the Surface Air Temperature hindcasts for the same region should be reasonable—at least for the purpose of this introductory post.
As I’ve illustrated in numerous earlier posts here, the long-term trend, since 1900, of the more commonly used NINO3.4 Sea Surface Temperature anomalies is basically flat. The same holds true for NINO3 Sea Surface Temperature anomalies, as shown in Figure 3. Based on the linear trend, there has been no rise in the Sea Surface Temperature anomalies for the NINO3 region since 1900. El Niño events dominated from 1900 through the early 1940s, La Niña events prevailed from the early 1940s to the mid 1970s, and then from 1976 to 2005, El Niño events were dominant. In other words, there is a multidecadal component to the frequency and magnitude of El Niño and La Niña events. That’s the reason a trend appears in the data that starts in 1950, Figure 2; the shorter-term data begins during a period when La Niña events dominated and moves into an epoch when El Niño events dominated.
So how well do the ensemble members and ensemble mean of the CCSM4 hindcasts of NINO3 Surface Air Temperatures compare to the NINO3 Sea Surface Temperature observations? Refer to Animation 1. Each of the six ensemble members and the ensemble mean are illustrated by individual graphs. The ensemble member graphs change every three seconds, while the ensemble mean remains in place for six seconds. (This also holds true for Animations 2 and 3.)
The first thing that’s obviously different is that the frequency and magnitude of El Niño and La Niña events of the individual ensemble members do not come close to matching those observed in the instrument temperature record. Should they? Yes. During a given time period, it is the frequency and magnitude of ENSO events that determines how often and how much heat is released by the tropical Pacific into the atmosphere during El Niño events, how much Downward Shortwave Radiation (visible sunlight) is made available to warm “and recharge” the tropical Pacific during La Niña events, and how much heat is transported poleward in the atmosphere and oceans, some of it for secondary release from the oceans during some La Niña events. If the models do not provide a reasonable facsimile of the strength and frequency of El Niño and La Niña events during given epochs, the modelers have no means of reproducing the true causes of the multiyear/multidecade rises and falls of the surface temperature anomalies. The frequency and magnitude of El Niño and La Niña events contribute to the long-term rises and falls in global surface temperature.
Of even greater concern are the NINO3 Surface Air Temperature linear trends exhibited by the CCSM4 model ensemble members and model mean. As discussed earlier, there has been no rise in eastern equatorial Pacific sea surface temperature anomalies from 1900 to present, yet the CCSM4 ensemble members and mean show linear trends that are so high they exceed the rise in measured global surface temperature anomalies. In the real world, cool waters from below the surface of the eastern equatorial Pacific upwell at all times except during El Niño events. It is that feed of cool subsurface water that helps to maintain the relatively flat linear trend there.
The trend in the NCAR CCSM4 NINO3 Surface Air Temperature anomaly hindcast is consistent with their hindcast of NINO3 Sea Surface Temperature anomalies from their previous version of the CCSM coupled climate models, which was the CCSM3. Figure 4 compares observed NINO3 Sea Surface Temperature anomalies to the hindcast of the CCSM3. (There is only one CCSM3 model run of Sea Surface Temperatures available through the KNMI Climate Explorer.) While the trend of the CCSM3 hindcast of NINO3 Sea Surface Temperature anomalies may not be as high as the trend of the CCSM4 hindcast of NINO3 Surface Air Temperatures, they are still showing a significant trend.
And to contradict this, the NCAR website presents NINO3.4 SST anomalies (They do not provide NINO3) with a flat trend over this period. Refer to Figure 5. So it appears as though NCAR understands that eastern equatorial Sea Surface Temperatures have not risen since 1900, based on the linear trend. Yet for some reason, their CCSM4 couple climate model cannot recreate this. (The data for Figure 5 is available at the NCAR webpage here. The dataset was prepared for the Trenberth and Stepaniak (2001) paper “Indices of El Niño evolution.”)
To answer the question that heads this section, the CCSM4 coupled climate model does a poor job hindcasting two important aspects of the El Niño-Southern Oscillation.
(For those new to my posts on ENSO, refer to ENSO Indices Do Not Represent The Process Of ENSO Or Its Impact On Global Temperature.It is written at an introductory level and discusses and illustrates with graphs and animations how and why El Niño and La Niña events are responsible for much of the rise in Global Sea Surface Temperatures over the past 30 years, the era of satellite-based Sea Surface Temperature data.)
HOW WELL DOES CCSM4 HINDCAST GLOBAL SURFACE TEMPERATURE ANOMALIES?
Meehl et al (2011) used HADCRUT Global Surface Temperature anomaly data in their Supplementary Information, so we’ll compare the HADCRUT Land plus Sea Surface Temperature anomaly dataset to the Ensemble Members and Mean for the Surface Air Temperatures of the NCAR CCSM4 on a Global basis. Refer to Animation 2. (It’s formatted the same as Animation 1: Observations Versus Six Ensemble Members and Model Mean.) The most obvious differences between the observations and the model outputs are the trends. The modeled trends are about 50% higher than those observed from 1900 to 2005. That’s a major difference. The other obvious difference is the CCSM4 ensemble members and mean do not appear to have the multidecadal component that is so apparent in the Global Surface Temperature anomaly records. Observed Global Surface Temperatures rose from the 1910s to the 1940s, dropped slightly from the 1940s to the 1970s, and then rose again from the 1970s to the late 1990s/early 2000s. The model outputs rise in the latter part of the 20th century, but fail to rise at a rate comparable to the observations during the early part of the 20thCentury and fail to drop from the 1940s to the 1970s.
And to answer the question that heads this section, the CCSM4 coupled climate model does a poor job hindcasting two important and obvious aspects of the Global Surface Temperature anomaly record from 1900 to 2005.
One of the known contributors to the multidecadal variations in Global Surface Temperature anomaly record is the mode of natural variability called the Atlantic Multidecadal Oscillation, or AMO. One might suspect that the AMO does not exist in the CCSM4. Let’s check.
HOW WELL DOES CCSM4 HINDCAST THE ADDITIONAL VARIABILITY IN NORTH ATLANTIC SEA SURFACE TEMPERATURE ANOMALIES?
Note that this is another portion of this post I will redo if and when the CCSM4 Sea Surface Temperature outputs are made available through the KNMI Climate Explorer. Also note that the Atlantic Multidecadal Oscillation is typically represented by detrended North Atlantic Sea Surface Temperature anomalies. But the multidecadal variations are easily visible in the “un-detrended” data, so I have not bothered to detrend it in the following graphs.
As discussed earlier, the Sea Surface Temperature outputs of the NCAR CCSM4 are not yet available through the KNMI Climate Explorer. But Sea Surface Temperature anomalies (detrended) are typically used to illustrate the multidecadal variations in the temperature of the North Atlantic. Again, like the global data, we’ll have to assume that the Marine Air Temperature outputs of the model mimic the Sea Surface Temperatures. The second concern is that land makes up 24% of the area included in the coordinates used for the North Atlantic (0-70N, 80W-0), as shown in Figure 6. The variability of land surface temperature can be different than that of Sea Surface Temperatures.
But as we can see in Figure 7, the instrument observation-based Sea Surface Temperature anomalies of the North Atlantic are tracked quite closely by the observed Land-Plus-Sea Surface Temperature anomalies of the North Atlantic “Plus” (where the “Plus” includes the additional Land Surface Temperature anomaly data encompassed by those coordinates).
The NOAA Earth System Research Laboratory (ESRL) uses a 121-month running-average filter to smooth their Atlantic Multidecadal Oscillation data. Refer to the ESRL AMO webpage. If we smooth the North Atlantic Sea Surface Temperature anomalies and the North Atlantic “Plus” Land+Sea Surface Temperature anomaly observations using the same 121-month filter, Figure 8, we can see the two curves are nearly identical.
So for the purpose of this post, the comparison of Land+Sea Surface Temperature anomalies to the Surface Air Temperature anomalies of the CCSM4 hindcasts will provide a preliminary look at whether there is a multidecadal component in the North Atlantic “Plus” data where one would expect to find it.
Animation 3 compares observed North Atlantic “Plus” Surface (Land+Sea) Temperature anomalies to the modeled Surface Air Temperatures for the 6 individual ensemble members and the ensemble mean. All data have been smoothed with a 121-month filter. Only two of the six ensemble members hint at multidecadal variability, but the frequency and magnitude are not comparable to the observations.
The NCAR CCSM4 coupled climate model appears to do a poor job of hindcasting the multidecadal variability of North Atlantic temperature anomalies.
NOTE ON MULTIDECADAL VARIABILITY OF MODELS
NOTE: Dr. Kevin Trenberth, Distinguished Senior Scientist at NCAR, and a lead author of three IPCC reports, provided a good overview of the models used in the IPCC AR4 released in 2007. Refer to Nature’s Climate Feedback: Predictions of climate post. There he writes:
“None of the models used by IPCC are initialized to the observed state and none of the climate states in the models correspond even remotely to the current observed climate. In particular, the state of the oceans, sea ice, and soil moisture has no relationship to the observed state at any recent time in any of the IPCC models. There is neither an El Niño sequence nor any Pacific Decadal Oscillation that replicates the recent past; yet these are critical modes of variability that affect Pacific rim countries and beyond. The Atlantic Multidecadal Oscillation, that may depend on the thermohaline circulation and thus ocean currents in the Atlantic, is not set up to match today’s state, but it is a critical component of the Atlantic hurricanes and it undoubtedly affects forecasts for the next decade from Brazil to Europe. Moreover, the starting climate state in several of the models may depart significantly from the real climate owing to model errors. I postulate that regional climate change is impossible to deal with properly unless the models are initialized.”
I suspect we’ll see a similar proclamation when AR5 is published.
Kevin Trenberth then tries to explain in the Nature.com article linked above why the differences between the observations and the models do not matter. But they do matter. When a climate change layman (one who makes the effort to look) discovers that the NCAR model CCSM4 hindcasts a global temperature anomaly curve that warms 50% faster than the observed rise from 1900 to 2005 (as shown in Animation 2), they question the model’s ability to project future global temperatures. The perception is, if the hindcast is 50% too high, then the projections must be at least 50% too high. And when the models don’t resemble the global temperature observations, inasmuch as the models do not have the multidecadal variations of the instrument temperature record, the layman becomes wary. They casually research and discover that natural multidecadal variations have stopped the global warming in the past for 30 years, and they believe it can happen again. Also, the layman can see very clearly that the models have latched onto a portion of the natural warming trends, and that the models have projected upwards from there, continuing the naturally higher multidecadal trend, without considering the potential for a future flattening for two or three or four decades. In short, to the layman, the models appear bogus.
A NOTE ABOUT MARINE AIR VERSUS SEA SURFACE TEMPERATURES
Sea Surface Temperature anomaly data are used in the GISS, Hadley Centre, and NCDC global temperature anomaly products. Yet as shown earlier, there are Marine Air Temperature datasets available. I used one, the Hadley Centre’s MOHMAT, in Figures 1 and 2. Sea Surface Temperatures are used for a number of reasons, some of which are discussed in Chapter 3 of the IPCC AR4. (24MB). (A word find of “Marine” or “NMAT”, without the quotes, will bring you to the discussions.) One of the reasons Sea Surface Temperature data is preferred is data availability. If you thought the global source data coverage for Sea Surface Temperature data was poor, there are even fewer instrument observations for Marine Air Temperature. Animation 4 illustrates a series of maps that indicate in purple which 5 deg by 5 deg grids contain data. It doesn’t indicate whether there are 30 observations, or 300, or 1 in a given month, just that there is data in a purple grid. The Animation 4 starts with January 1900 and progresses on a decadal basis through January 2000.
As illustrated, Marine Air Temperature observations in the Southern Hemisphere are rare south of 30S before 1950, they’re rare globally for that matter in the first half of the 20thCentury, and data is virtually nonexistent north of 60N in the Northern Hemisphere even through 2000.
Based on that, we’ll limit the comparisons of observed and modeled Marine Air and Sea Surface Temperature data for the global oceans to 30S-60N, and start them in 1950. The end month of December 1999 is dictated by the hindcasts of the NCAR CCSM3, which is the earlier version of that NCAR coupled climate model. Also note that there was only 1 model run for the CCSM3 Sea Surface Temperatures at the KNMI Climate Explorer.
Figure 9 compares the linear trends of a Marine Air Temperature anomaly dataset (MOHMAT) to two Sea Surface Temperature datasets (HADSST2 and HADISST) for the latitudes of 30S to 60N, from January 1950 to December 1999. These are instrument observation-based datasets. As illustrated, the Marine Air Temperature anomalies rise at a rate that is significantly less than the two Sea Surface Temperature anomaly datasets. The linear trend of the Marine Air Temperature anomalies is about 52% of the average of the trends for the two Sea Surface Temperature anomaly datasets.
On the other hand, the modeled ensemble mean for the Marine Air Temperature output of the NCAR CCSM3 (earlier version) has a linear trend that is more than double the trend of the modeled Sea Surface Temperature anomalies. The relationship is backwards. Does this backwards relationship between Sea Surface and Marine Air Temperatures continue to exist in the CCSM4?
The preliminary look at the hindcasts of the NCAR CCSM4 sheds a different light on the model-based papers of Meehl et al (2011) and Stevenson et al (2011). Those papers are based on a coupled climate model that cannot reproduce essential portions of the 20thCentury Surface Temperature observations.
No matter how well the NCAR CCSM4 can simulate certain aspects and processes of global climate, the fact that it cannot reproduce many portions of the instrument temperature record during the 20thCentury emphasizes failings that call into question its ability to project future global or regional climate change.
All observation-based data presented in the post are available through the KNMI Climate Explorer Monthly observationswebpage, with one exception.
The NCAR NINO3.4 data used in Figure 5 is available through the NCAR TNI (Trans-Niño Index) and N3.4 (Niño 3.4 Index) webpage.