Guest Post by Willis Eschenbach
[UPDATE: An alert commenter, Ken Gregory, has pointed out that in addition to the temperature affecting the CRE, it is also affected by the changing solar radiation. He is correct that I did not control for this. SO … I need to go off and re-think and then re-do the entire analysis. In the meantime, in the immortal words of RMN, my analysis below is no longer operative. Bad Willis, no cookies … but that’s the nature of science. Thanks, Ken, for pointing out my error. -w.]
[UPDATE: See the subsequent post here. -w.]
Figuring that it was about time I did some more scientific shovel-work, I downloaded the full ten-year CERES monthly satellite 1° x 1° radiation dataset (link below). I also got the Reynolds monthly Sea Surface Temperature 1° x 1° dataset, and the GHCN monthly 1° x 1° land dataset. This gave me nominally complete ten-year gridded data for the ten-year period from March 2000 through February 2010 for both the temperature and the radiation.
Among the CERES datasets are the shortwave-, longwave-, and net- cloud radiation effect (CRE). Clouds affect the radiation in a couple of ways. First, clouds reflect sunlight so they have a big cooling effect by cutting the downwelling shortwave radiation. In addition, however, they are basically perfect blackbodies for longwave radiation, so at the same time, they warm the surface by increasing the downwelling longwave radiation. And of course, at any instant, you have the net of the two, which is either a net cooling effect (minus) or a warming effect (plus). All of these are measured in watts per square metre (“W/m2”).
So without further ado, Figure 1 shows the net cloud radiative effect (CRE) from the ten years of CERES data. It shows, for each area of the earth, what happens when there are clouds.
Figure 1. Net cloud radiative effect (CRE). Red and orange areas show where clouds warm the earth, while yellow, green, and blue show areas where clouds cool the earth. The map shows that if there is a cloud at a certain area, how much it will affect the net annual radiation on average.
Note that in some areas, particularly over the land, the net effect of the clouds is positive. Overall, however, as our common experience suggests, the clouds generally cool the earth. But this doesn’t answer the interesting question—what happens to the clouds when the earth warms up? Will the warming cloud feedback predominate, or will the clouds cool the earth? It turns out that the CERES data plus the earth temperature data is enough to answer that question.
What I’ve done in Figure 2 below is to calculate the trend for each gridcell. The meaning of the trend value is, if the surface temperature goes up by a degree, what do the clouds do to the radiation? I used standard linear regression for the analysis,. It’s a first cut, more sophisticated methods would likely show more. As is always true in the best kind of science, there were a number of surprises to me in the chart.
Figure 2. Slope of the trend line of the net cloud radiative effect as a function of temperature. This give us the nature of the cloud response to surface warming in different areas of the world. This is what is commonly known as “cloud feedback”, although it is actually an active thermoregulatory effect rather than a simple linear feedback.
The first surprise to me is the size of the variation in cloud response. In some areas, a 1° rise in temperature causes 20 extra W/m2 of downwelling energy, a strong warming effect … and in other areas for each 1° fall in temperatures, you get the same 20 extra watts of downwelling energy. I didn’t expect that much difference.
The second surprise was the difference in the polar regions. Antarctica itself is cooled slightly by clouds. But when temperatures rise in the Southern Ocean around Antarctica, the clouds cut down the incoming radiation by a large amount. And conversely, when the temperatures in the Southern Ocean fall, the clouds provide lots of extra warmth. This may be why the Antarctic and Arctic areas have responded so differently to the overall slight warming of the globe over the last century.
The third surprise was the existence of fairly small areas where the cloud response is strongly positive. It is surely not coincidental that one of these is in the area of the generation of the El Nino/La Nina events, near the Equator on the west side of South America.
One thing that did not surprise me is that the reaction of the clouds in the area of the Inter-Tropical Convergence Zone (ITCZ) in the Pacific. This is the greenish band about 10° North of the Equator across the Pacific and across the Atlantic. In this area, as I’ve shown in a variety of ways, the cumulus clouds strongly oppose the rising temperature.
Finally, there’s one more oddity. This is the fact that overall, as an area-weighted average trend, for every degree the globe warms, the warming is strongly opposed by the cloud radiation effect. The action of the clouds reduces the downwelling radiation by 3 W/m2 for every degree the planet warms … in IPCC terminology, this is not only a negative feedback, but a strong negative feedback.
And the cooling effect of the clouds is even stronger in the ITCZ. There, for every degree it warms, the downwelling radiation drops by ten W/m2 or so …
I think, although I’m by no means sure, that this is the first global observational analysis of the size of the so-called “cloud feedback”. It shows that the cloud feedback is strongly negative overall, -3 W/m2 for each degree of warming. In addition, in the critical control areas such as the ITCZ, the cooling effect is much larger, 10 W/m2 or so. Finally, it shows a very strong negative cloud feedback, 20 W/m2 or more, in the area of the Southern Ocean
Like I said … lots of surprises. All comment welcome, and please remember, this is a first cut at the data.
w.
DATA
Land Temperature Data: From KNMI, in the “Land” temperature section, identified as the “CPC GHCN/CAMS t2m analysis 1.0°”.
Sea Temperature Data: Again from KNMI, in the “SST” temperature section, identified as the “1° Reynolds OI v2 SST, v1”.
Once you click on the observations you want, at the bottom of the succeeding page is a link to a NetCDF (.nc) file containing all of the data.
CERES Data: From NASA (offline now, likely the Gov’t shutdown), identified as “CERES_EBAF-TOA-Terra_Ed2.5_Subset_200003-201002.nc”
If you don’t want to mess with the underlying datasets, I have collated the CERES and the temperature datasets into a series of arrays in R, that are 180 row x 360 column x 120 layers (months) in size. They are available here, along with the corresponding arrays for the surface temperatures, and a landmask and a seamask file. WARNING—Be aware that this is a large file (168 Mb).
The file is an R “Save()” file named “CERES long”, so it is loaded as follows:
> mytest=load("CERES long")
> mytest
[1] "toa_sw_clr" "toa_sw_all" "toa_lw_clr" "toa_lw_all" "toa_net_clr" "toa_net_all" "cre_sw" "cre_lw" "cre_net" "solar" "landmaskarr" "seamaskarr" "allt"<
In the naming, “toa” is Top Of Atmosphere, “sw” is shortwave, and “lw” is long-wave; “all” is all-sky, “clr” is clearsky; “cre” is cloud radiative effect, “solar” is downwelling solar”, and “allt” is all the temperature records (land and sea).
The R program I used is here … but I must warn you that far from being user-friendly, it is actively user-aggressive. Plus it has lots of dead code. Also, none of my programs ever run start to finish, they are run in chunks as needed. However, the functions work, and the mapping section (search for “MAPSTART”) works.
Some ask question “How can clouds make it warmer”. This is a matter of how much warming you have in the sun. If you live where it gets cold in winter, you learn it. Clouds make it warmer when sun in not warming much (winter or nights), but colder when sun is warm (summer and day).
— Mats —
Very nice, my only suggestion is tweaking the palette on the map and getting this published! Of course it would be completely invisible to the IPCC just like the Terra Satellite data and study by Spencer but it needs done regardless.
I was going to suggest the same thing that I can’t believe they haven’t had a look. They probably have and got some inconvenient results. Such results cannot and will not be allowed to pass into the IPCC reports.
Without any research on my part whatsoever I conclude that clouds are a net negative feedback. Take a look at 60 millions years ago down to the present day, look at the changes in co2 and “inferred” cloudiness, we should have our goose cooked by now.
One can also see why ‘The IPCC’ have such trouble with Clouds. Not only do they reflect SW and interfere with the ‘orderly exit’ of LW … The LW could be from picked up from anywhere in their path, not just the surface that they are currently over. There is also the element of their initial energy (measured as they coalesce and dump it) having been ‘picked up’ when they just molecules leaving the Sea on the Equator (for example).
Hmm … glad I don’t have to model that one.
.
AntonyIndia says:
October 3, 2013 at 11:35 pm
I’m not finding that … my analysis shows that there is a very tiny decrease, just under half a watt per square metre, over the decade of the data. The decomposition of the CERES OLR data is as follows:
As you can see, the annual average (“Trend”) varies by only about 0.8 W/m2 over the period and is fairly irregular. I wouldn’t put much stock in a purported “decrease” in OLR.
w.
In addition. It would seem that as long as that graph you presented ends at the same (24/0 hour) point each day then the net effect of cloud over 24 hours is … zero?
I am stunned that this has not been done before. It seems to me that this ought to have been done Monday Morning 8:00 am in the 1960s.
I am actually stupefied that this isn’t just something that we know. The fact that Willis can suggest a negative feedback and it bears out in the data is a shock to me. Has this been overlooked?
Or buried?
Cloud altitude, cloud composition, cloud cover thickness, multiple layers of different types of clouds and day vs night clouds, it would seem should all have differing effects upon the resultant temperature.
RC Saumarez says:
October 4, 2013 at 1:31 am
Thanks, RC. The CERES data are monthly averages, not monthly samples. The CERES satellite images a strip around the earth with every revolution. All of these are collated, gridded, and averaged.
w.
Willis,
Forgive me if this is covered, but I do not see a mention of the effect of clouds on downwelling IR.
It seems to me that any IR emitted at high altitude from clouds, water vapor or CO2 would be more likely to penetrate the thin, dry atmosphere above than the thickening, moist atmosphere below. If IR is radiated in random directions, then the net flow would be toward space due to the resistance of the thicker, moister atmosphere below. (simple terms, but trying to illustrate a concept I do not see much).
Willis, do you have a link to your first post on all this where you had sat photos of cloud spreading across tropics in relation to time of day. From memory that was a fairly thin line along ITCZ. That was quite an ingenious demonstration of the effect that I found quite convincing. It would be interesting to compare to these maps.
Great stuff Willis. I really need to learn R.
So this is all 10 year average info? If so, it’s amazing that you still have that much signal with so much smearing going on. I’m sure you’ll delve into the dynamics (it’s something I’ve been dabbling with), and I think you will see truly GIANT cloud effects if you study one cell during the time when you know very large storms happened there.
For example, we know that strong rain at 1″ per hour will give up about 15,924 W/m^2 due to latent heat of vaporization. (distributed somewhere in the column above that 1m^2). So 1″ of rain is like leaving 10 space heaters in 1 square meter on full blast for an hour. So the effects happening are truly huge. Warmed air is carried aloft, and it stops when the lift equals the surrounding buoyancy. And it spreads out. The next cool thing is that the water droplets on the top side of the cloud (if clear above) have now been transported high enough that there simply isn’t much GHG above to scatter IR, so you’ve effectively shunted surface warmth to space with very little GHG interaction. You simply bypassed the vast majority of the GHG effect. As you have pointed out, the stronger the storm, the more +feedback in strength it has (until it exhausts its fuel), the higher it punches through the atmosphere, and the stronger the ability to radiate more directly to space is. So this outer layer is going to radiate away its energy very quickly to space, and even more quickly the higher the altitude (with the offsetting change in radiation rate due to temperature/altitude effects). Subsequent layers of water droplets are also going to radiate, but at warmer temperatures because there are water droplets above that which are also radiating in all directions. These lower layers have a reduced ability to reject heat to space, but you’ve also just created a thermal differential (cooler top loses energy at a higher rate than layer below) so you also now have a convective engine to keep bringing warmer droplets to the top where they radiate out faster. Obviously, the water droplets can’t radiate at 15,924 W/m^2, but to the extent that they do radiate (SB, T^4), the ratio will largely be a factor in determining the area that this top layer (the anvil) will occupy, which also determines the (also huge) effect of reflected sunlight. The layer thickness at which radiative energy transfer gets cut in half inside the cloud is also an important metric in determining the temperature differential, propensity to form thermals, and resulting strength of convective processes to help transfer the heat to the outer radiating surface.
I think you’ll find short term effects in these datasets that will confirm what you’ve been saying all along, that the ability to reject heat on demand is so staggeringly huge, that whatever small effect that CO2 may produce is simply caught up in the dragnet and rejected with all of the other huge swings we see on a daily basis. 15,924 W/m^2 > 2 W/m^2. Of course, the grid size is still way too coarse to uncover the true scale of these effects. The true magnitudes of these factors will be far higher than can be seen in a 1°x1° grid since thunderstorms almost never get that large.
This sounds like such a blast to work on. I’m convinced you are working on the very issue that deals with why climate models fail (and always run hot). Heat doesn’t collect here (and go “missing”), it gets actively rejected on demand by an enormously powerful heat shunting engine.
I can even imagine a cartoon with the first image of earth from space. Willis comments from his spaceship “Huh, Look! Textbook example of a vapor phase change thermostatically governed planet!”
Where I live cloud in winter means warming while cloud in summer means cooling. To avoid paradoxical (in the statistical sense) misinterpretation of misleadingly aggregated stats (pooling across key conditional lurking variables), better next extend the exploration to seasons. Also recommend (to anyone exploring further) constructing a battery of diagnostic multivariate scatterplot matrices to look for things like residual kinks & heteroskedasticity changes near 0 degrees C (due to phase change of water) and sensitive regional dependence of regression coefficients on start & end dates due to spatiotemporally nonrandom interannual variation (since the record length is only 10 years). Map animations (by time of year) of the sensitivity of regression coefficients could point to regions where standard mainstream assumptions about equator-pole transports fail catastrophically. Deepening insight via due diagnostics — i.e. learning from systematic patterns of regression model assumption failures by looking carefully at nonrandom residual patterns in every which way possible, always remaining vigilantly, lucidly, & consciously aware that regression coefficients depend conditionally on the absence of omitted key conditioning variables. Based on the looks I’ve had at these multivariate relations, I would not recommend naively trying to derive a single coefficient that applies globally at all times — careful diagnostics strongly counsel against this. Those who have the time & resources can explore every path to everywhere, but for those short on luxuries there are universally-constrained paths that afford more efficient & decisively-conclusive exploration.
“ So without further ado, Figure 1 shows the net cloud radiative effect (CRE) from the ten years of CERES data. It shows, for each area of the earth, what happens when there are clouds. ”
*choke* *choke* … cough cough … can’t take a bite of that w/o a reaction … if this was meant as an IQ test (‘trap’) on the general readership here, you succeeded; I don’t think the graphic depicts quite what has been expressed …
.
Willis,
Interesting post; I especially like the graphics. One doubt comes to mind: The correlation is very strong, but does that necessarily imply causation? For example, in the eastern tropical pacific, it is clear that warmer water (El Nino) is associated with greater cloudiness (higher ocean surface temperature, more water vapor, more clouds), so your graphic shows strong ‘positive cloud feedback’ (suggesting more clouds warm the water in the eastern tropical Pacific), when it is pretty clear based on ENSO that the causation is the other way around (warming eastern tropical Pacific causes more clouds). I suspect that the same cause/effect uncertainty is associated with much of the correlation your graphic shows. Similar analyses have appeared in the literature (Lindzen, and then Lindzen and Choi, along with papers claiming to show the Lindzen & Lindzen and Choi correlation is not causal). The key seems to be an adequate lead/lag analysis of the correlation, but IIRC, even that is not always enough to give a clear answer.
Willis is over reaching with his research once again,(as was done with his volcanic study).
Therefore I say interesting but take it with a grain of salt.
And, as ever, we are waiting for you to be SPECIFIC. What EXACTLY is Willis ‘over reaching# with – could you clarify.
Or are you yet another idiot who comes here in order to demonstrate their unique ability to pollute a post with crap. Seriously, you follow Willis like a rash, PUT UP OR SHUT UP FFS.
(BE SPECIFIC. What EXACTLY..,)
tonyb says:
October 4, 2013 at 1:59 am
Upon further research, I find there are a couple of papers out there. One uses MODIS data, the other CERES. Both are paywalled, one here and the other here.
Both find different answers (one positive, one negative). They are much smaller than what I find. I suspect that this is because they’ve (wrongly in my estimation) used yearly averages rather than actual data. Don’t know why. I’ve checked my results, I stand by them (until errors are found, of course).
w.
clivebest says:
October 4, 2013 at 2:49 am
Of course I’ve used the seasonal data. Why would you want to average it and lose all of the actual changes?
What you call “seasonal” changes are changes due to the ground warming and cooling. This gives us the perfect way to see what actually happens to the clouds when the earth actually warms and cools.
And what we find is that as a global average, for every degree warmer that the earth gets, the downwelling radiation drops by about 3 W/m2. Now, that’s just facts. How you want to interpret that is up to you … but as the earth warms and cools, that’s what happens.
w.
I am a bit confused. I would have guessed that the areas positively correlated temperatures with cloudiness would be dominated by nighttime cloudiness, and areas of negatively correlated temperatures in areas dominated by daytime cloudiness (such as the tropics)
cd says:
October 4, 2013 at 8:42 am
Steven Mosher
Would you entertain these questions. I’m sure they are misconceived but a short response to each would be appreciated.
Can you tell us why using the Kriging methodology, the BEST team do not provide the Kriging varaince maps. Is this because it is not forthcoming in your implementation; or is it just that it highlights more uncertainty; or do you express this as part of your confidence in the final time series?
###############
See the appendix to the methods paper. Posting the spatial uncertainty in grids is on the to do list. For every update we generate over 300K files. plowing through those is quite a job, but
once I’m sure that is operating correctly we’ll move on to added more data.
########################################
As I remember the implementation assumed a “geomodel” of spatial temperature trends in order to derive the residuals. This seemed somewhat contrived to me. Why was this better than say using radial basis functions of the raw unprojected data or B-Spline of the projected raw data to compute your 0 plane.
hmm. you dont understand the process. The temperature at any given location is expressed
as a deterministic function and a random component. read the appendix to the methods paper
and then read around in the kriging literature. the approach is standard cookbook. There are some splining approaches that could also be substituted.
Willis –
I can tell you right now that the map is wrong for where I live – the dead center of Mexico, in Mexico’s Central Valley. That area on your map is shown orange, indicating that clouds warm.
That is not true at all. Whatever you’ve done to derive that map, it doesn’t hold true here.
When clouds come the area gets about FIVE degrees C cooler.
Our climate here is VERY stable, throughout the year – maybe the best and most consistently flat climate in the world. The hottest to coolest high temps only range about 20°F(85°F to 65°F), summer to winter. And the reality is that when clouds come, it INVARIABLY cools off.
Speaking of clouds…
Decades ago, for reasons and purposes I really can’t discuss, I was looking into the vertical tranmissivity of the atmosphere with respect to a certain band of the infrared spectrum. In our problem, clear skies were okay, but clouds were regarded as complete blockage. We assembled the statistics for cloud cover and were quite surprised to discover that most of the Earth, on a time-averaged basis, had prevailing cloud cover. We were seeking cloud-free lines of sight from stations on the ground, and were narrowly constrained. The problem was so bad, we generally concluded that a system solution using that approach was the last resort.
I can’t recall whether the subject of worldwide cloud coverage statistics has been discussed, but if it hasn’t, it would be worthwhile.
People interested in this topic should take a good long look at the work of Roy Spencer, starting with his blog at http://www.drroyspencer.com. From there you can go on to his technical papers and/or his book “The Great Global Warming Blunder”.
Most of the mainstream climate models (at least at the time of AR4) took note of observations (made quite a while ago) that showed lessened cloud cover correlated with warmer conditions, then made the assumption that warming conditions caused reduced cloud cover. As a result, virtually all of the models used in AR4 (I have not had time to check AR5) show a positive cloud feedback effect.
Several years ago, NASA GISS scientists led by Andrew Lacis published an article in Science
http://www.sciencemag.org/content/330/6002/356.short
where they used the GISS climate models to simulate what would happen if all CO2 were removed from the atmosphere. The model produced a virtual runaway that led to a 90% reduction in atmospheric water vapor (from it condensing out of the cooling atmosphere) and also a 50% increase in low cloud cover so that over three-quarters of the world would always be clouded over. A desert planet perpetually shrouded in clouds – how plausible!
Anyway…
Since reducing cloud cover due to some other means would lead to warming (direction of causality going the other way), this assumption should be checked very carefully, as Roy has been advocating. I don’t know if Roy’s conclusions or inclinations are correct, but the bigger issue to me is that no one else in the climate establishment has really looked at this issue with the detailed attention it deserves.
“Since reducing cloud cover due to some other means would lead to warming (direction of causality going the other way), this assumption should be checked very carefully, as Roy has been advocating.”
Indeed this seems to have been happening. About half the warming from 1980 to 1999 can be attributed to a reduction in cloud cover during this period. After 1999 cloud cover then stabilized before increasing slightly. Combined with CO2 forcing this can explains both the rapid warming prior to 1999 and the consequent hiatus. As a result CO2 climate sensitivity is reduced. Euan Mearns and myself have a paper describing this effect which is under consideration.
Very interesting post. I’m sure there are many ways to look at this.
Ever since AR4’s admission of lousy understanding of clouds and the blithe statement that they were generated randomly in climate models, I have
a) thought of this as the achilles heel of the catastrophic warming hypothesis
b) wondered what weather conditions (including temperatures in the oceans) lead to consistent biases towards daytime or nighttime clouds.
Perhaps this dataset and some old-fashioned meteorology can answer (b)?