Do clouds control temperature, or does temperature control clouds?

Previous Energy Matters posts that have dealt with the cloud/temperature relationship (here and here) have concluded, as have most other studies, that cloud cover acts as a control on global surface temperatures. In this post I dust off a widely ignored but apparently robust cloud cover series – the ICOADS ocean series – which suggests that ocean surface temperatures may in fact be acting as a control on cloud cover rather than the other way round.

In his 2015 post on NASA climatology data Euan Mearns presented a version of this graph:

Figure 1: ISCCP cloud cover vs. HadCRUT4 surface temperature

And captioned it thus:

ISCCP global mean cloud amount in black varies inversely with temperature in blue. No major surprise there, less cloud and a warmer surface, but what causes what?

What causes what indeed? It is of course commonly assumed that fewer clouds cause warmer temperatures and more clouds cooler temperatures. But how you interpret the relationship depends on which cloud data set you use, and here I present some data sets that suggest an alternative interpretation.

The ISCCP series is one of a number of cloud cover series derived from data collected by the AVHHR, MODIS and other satellites, and Figure 2 shows all of them plotted together. The conclusion reached by the World Climate Research Programme, which included this graph as Figure 3.5.1 in its voluminous 2012 report, was that you can’t deduce anything from these results: “At present, one can only conclude that global mean monthly cloud amount is constant over the last 25 years to within 2.5%, within the range of annual variability”:

Figure 2: Global cloud cover data derived from all satellites

Here’s a simplified version of the Figure 2 plot from Nigel Calder’s website:

Figure 3: Selected satellite and other global cloud cover series

The ISCCP cloud cover series seems to be cherry-picked, which leaves me free to cherry-pick a series of my own. I’ve chosen PATMOS-x, which is actually a little longer than ISCCP. Figure 4 isolates it.

Figure 4: PATMOS-x cloud cover series

Figure 5 now adds four more series – scales are not shown partly because I don’t know how to make them compatible but mostly because the idea is simply to show trends.

The first series is one that hardly anyone ever looks at – the ICOADS ocean cloud cover series, which I downloaded from KNMI Climate Explorer. Note that because the oceans are cloudier than the land roughly 80% of the world’s clouds form over the oceans.

The second is the NASA NVAP-M atmospheric water column series, which we might expect to be related to cloud cover. (h/t to Clive Best for bringing it to my attention):

The third is the unadjusted ICOADS SST series, also downloaded from KNMI (HadSST3 is essentially the same over this time period).

The fourth is a crude measure of the average tropospheric lapse rate that I put together by subtracting the RSS lower stratosphere temperature anomalies from the ICOADS SST anomalies. (The two downward excursions are present in the stratosphere record and were caused by the El Chichón and Pinatubo eruptions). Presumably lapse rate would also have some impact on cloud formation, although I don’t insist on it:

Figure 5: Five cloud-cover and cloud-cover-related series

All five series show the same basic trends – increasing before the late 1990s, peaking around the time of the 1997/8 El Niño and deceasing or flattening thereafter.

This congruence of trends seems a little too close to be coincidental. It implies a relationship between temperature and clouds in the sense opposite to that usually assumed, i.e. with temperatures increasing rather than decreasing as cloud cover increases. It’s difficult to see how an increase in cloud cover could cause warming, so the supposition would be that the ocean surface warming is causing the increase in cloud cover.

The series shown in Figure 5, however, go back only to 1982. Can we trace a temperature/cloud relationship before then? Indeed we can. The ICOADS ocean cloud cover series, which extends back to 1855, shows a positive relationship with ICOADS SST (also downloaded from KNMI) going all the way back to around 1890 before it begins to break apart:

Figure 6: ISCCP cloud cover vs. HadCRUT4 surface temperature

If this relationship is real it raises the possibility that the +/-1C increase in ocean surface temperature during the 20th century caused a +/-10% increase in ocean cloud cover. The objection that the ICOADS cloud cover series, which is put together from visual shipboard observations, and the SST series aren’t reliable will inevitably surface here, so we will now address the question of whether they are reliable or not. The conclusion is that while their accuracy can’t be guaranteed there are no reasons to suppose that there is anything seriously wrong with them after 1960.

The question of the reliability of the ICOADS ocean cloud cover series, which is constructed from many millions of visual cloud cover observations made from ships, has been reviewed by various investigators, including:

Hahn et al. 1992: The use of COADS ship observations in cloud climatologies

Eastman 2011: Variations in cloud cover and cloud types over the ocean from surface observations, 1954-2008

Smith 2011: Oceanic Cloudiness Reconstruction Based on ICOADS (1900-2010)

And apart from some localized biases and a minor “night-reading” bias (observers tend to measure fewer clouds at night) none of them seems to have found anything seriously wrong with it. Smith sums it up thus: “The annual analysis indicates oceanic cloudiness variations that are broadly consistent with known climate modes such as ENSO. Variations also indicate multi-decadal variations, including trend-like variations over the 20th Century consistent with changes in global SSTs. Because of those consistencies with independent analyses it is likely that most of the analyzed variations are caused by climate variability.” Not a completely clean bill of health, but close.

The “trend-like variations over the 20th Century consistent with changes in global SSTs” that Smith refers to are seen in Figure 6 above. There is a strong correlation between ocean cloud cover and sea surface temperature (R^2 = 0.84 for annual means) after 1890. It persists during World War II, when both records show a large spike caused by a wartime observational bias, and also during the period of no warming since 1998, when both cloud cover and SST decrease or flatten out:

Why should ocean cloud cover and SST track each other over a period of 120 years? There are two possibilities. Either the SST data or the cloud data, or both, are distorted by biases that make them appear to be related, or they really are related.

We will deal with the bias issue first. Here’s a summary of the evidence for the existence of biases in the ICOADS cloud cover and SST records:

1. The SST/cloud correlation breaks down before 1890, raising the question of whether the correlation after 1890 might not be spurious. (An alternative explanation is bad and/or insufficient data before 1890.)

2. According to conventional wisdom the SST record is variably distorted by instrumental biases before about 1960 (I think it isn’t, but that’s beside the point). However, it’s also generally accepted that it isn’t seriously distorted after 1960.

3. Increasing areal coverage during the second half of the 19th century and the first half of the 20th could have caused an overall shift of observations from cooler and less cloudy to warmer and more cloudy areas, thereby making it appear that global SST and cloud cover were increasing even though they weren’t. Since 1960, however, coverage has been essentially stable except for the area south of 40S latitude, and deleting the small amount of data below 40S makes no significant difference to global cloud cover and SST trends (Figure 7).

4. Before 1950 there’s an inverse relationship between cloud cover estimates and the number of observations (Figure 8), which is evidence for a bias of some kind (the same relationship is visible in the SST record, but less well-defined). There is, however, no clear relationship between cloud cover or SST and the number of observations after 1950. (Figure 8 shows negative numbers of observations because KNMI treats nobs as a climate variable and expresses it as anomalies relative to the mean, but this doesn’t change the shape of the plot.)

5. According to Eastman et al. “a major change in the classification of cloud types (from tenths to oktas) was instituted in 1949, and it was not until 1954 that the change was adopted by ships of all nations”, which is why the Eastman reconstruction begins in 1954. There’s no evidence that this change distorted the cloud cover record but we’ll take it into account anyway.

6. The data during WWII are clearly bad and should be deleted.

Figure 7: ICOADS cloud cover observations, global coverage 1860-1960

Figure 8: ICOADS cloud cover series vs. number of observations

But none of these six potential sources of bias affects the data after 1960, so we can eliminate all of them simply by deleting the data before 1960. This leaves us with a continuous 64-year-long ocean cloud cover series that again broadly tracks sea surface temperatures:

Figure 9: ICOADS cloud cover vs. SST series, 1960 on

A few more checks before proceeding to see if we can detect any other flaws in the ICOADS records. Figure 10 matches the ICOADS SST data against independently-measured GISS  surface air temperatures since 1960 (data again from KNMI):

Figure 10: ICOADS SST vs. GISS surface air temperature series

There are no reliable independent data sets to compare the ICOADS cloud cover data against, but Figure 11, which compares monthly cloud cover records from 11 coastal stations in Europe (from the ECA data base) against the cloud cover data in the ICOADS grid blocks that overlap these stations, shows generally excellent agreement between land-based and shipboard observers over a period of almost 50 years:

Figure 11: Cloud cover observations, ICOADS shipboard vs. ECA land-based

And for what it’s worth Figure 12 shows a respectably close match between the ICOADS and ISCCP grid blocks covering these 11 stations after 1989:

Figure 12: Cloud cover observations, ICOADS shipboard vs. ISCCP satellite

From Figures 11 and 12 we can deduce that ground and shipboard observers do a good job of counting clouds.

Inevitably, however, there are instances where the cloud/SST correlation breaks down. Figure 13 plots ICOADS SST against ICOADS total cloud cover in the 30-60N, 0-30N, 0-30S and 30-60S latitude zones. The temperature-cloud correlation is present in the 0-30N, 0-30S and 30-60S bands but not in the 30-60N band:

Figure 13: ICOADS cloud cover and SST in different latitude zones

And while there’s a long-term correlation between global SST and global cloud cover in most places there is no correlation over shorter periods. Figure 14, which compares detrended monthly ICOADS global SST with detrended monthly ICOADS global total cloud cover, has an R^2 of 0.02.

Figure 14: Detrended ICOADS cloud cover and SST data

There is just one place where we see a strong short-term positive correlation between SST and cloud cover – in the El Niño zone (5N-5S, 160E-100W, Figure 15). Note that the two plots use the same Y-scale. One gets the impression that ENSO events, in particular strong El Niños, manufacture a lot of the world’s clouds:

Figure 15: ICOADS cloud cover (orange) and SST (blue) inside & outside El Niño zone

That concludes the data review. It isn’t fully diagnostic, but there is no good evidence that there is anything seriously wrong with either the ICOADS SST or the ICOADS ocean cloud cover series at least after 1960. That being the case I submit that the possibility that temperatures control clouds, and not the other way round, should be considered.


This entry was posted in Climate change and tagged , , , , , , . Bookmark the permalink.

15 Responses to Do clouds control temperature, or does temperature control clouds?

  1. michael says:

    From the analysis it would seem that there is insufficient data to say one way or the other. Temperature does increase or decrease evaporation of water and to form a cloud one needs water vapor and something for it to condense on. To cut to the heart of the matter how about a feed back loop. higher temperature more clouds, more clouds less solar gain and a lowered temperature so we have a thermostat with a long period time base.

  2. Cy Halothrin says:

    Prior to the 1960s (when weather satellites started to be launched) I don’t think it’s very easy to discern how cloudy it was relative to the present time. There may be a few million written weather reports that date back a couple of centuries, but it’s really hard to sift through that data and make any accurate statements about it.

    Now that we have weather satellites and computers to crunch the data, it should be easier to see what the trends are.

    One issue that “clouds” the data so to speak is the phenomena of “global dimming”

    As I understand it, this is less of an issue now that coal burning power plants are equipped with pretty effective smog “scrubbers.” I recall reading how back in the ’60s cities like Pittsburgh were so smoggy that you could barely see across the street, but that has been cleaned up. However, I’ve seen the same thing in China during the 1990s (the city of Benxi was memorable for that). I think that China has also been cleaning up its act in this regard, but I can’t say just where we stand right now in terms of particulate and sulfur dioxide pollution as compared to the past.

    There have also been a few exceptional years like 1991 when Mt Pinatubo erupted, sending massive amounts of sulfur dioxide into the atmosphere, enough to dramatically affect global temperatures.

    • Now that we have weather satellites and computers to crunch the data, it should be easier to see what the trends are.

      I think Figure 1 shows that satellites have so far not done a good job of counting clouds even after computer-crunching.

      One issue that “clouds” the data so to speak is the phenomena of “global dimming”

      According to your link global dimming occurred between about 1960 and 1990. Could it have been caused by an increase in cloud cover? Figure 9 suggests it could.

  3. Great post, Roger. I cannot really comment on technical/math matters but will try to add my 2c.

    a) It’s surprising that cloud cover and temperatures would be positively rather than negatively correlated. iirc most people who look into the issue find the opposite, most recently this paper.

    I don’t really think the paper shows anything other than the fact that hotter months/years have less cloud cover. They don’t address the cause-and-effect issue (is the water hotter because there are less clouds, or are there less clouds because the water is hotter).

    I haven’t really looked into the paper in detail, but the difference between their and your results could be due to:
    1: not using the observational cloud datasets you use here (they infer cloud cover from humidity)
    2: some oddity related to the way they measure correlation

    b) You say:
    ‘If this relationship is real it raises the possibility that the +/-1C increase in ocean surface temperature during the 20th century caused a +/-10% increase in ocean cloud cover.’

    While such a relationship can exist the ratio seems off. Clouds albedo is about 70w/m2 while their greenhouse effect is 30w/m2. If 80% of clouds are over the ocean, this suggests a 10% increase in cloud cover would cause a net increase of 3w/m2 in the energy budget. That seems way too high – it would cancel all the manmade forcing over the period.

    I’d also like to link to the most instructive and clear paper I’ve seen on this issue:

    The conclusion is a frustrating ‘we cannot tell yet’: cloud feedback might be positive or negative depending on the time period and dataset used.

    c) The previous commenter mentioned a ‘stabilization loop’ by which reduced cloud cover could cause increased temperatures which in turn increase cloud cover, which reduces temperatures again. Well something like that is what happens with isoprene: it’s released when sunlight makes contact with the ocean’s surface and can contribute to cloud formation.

    The problem is this was discovered only a few months ago and so we have no idea how much or little ocean-borne isoprene could affect global cloud cover.

    One suggestion. If you have the time, perhaps you could compare the data on cloud cover from the first paper I linked to with the observational datasets you show here. As I said, that paper infers that dry regions (in the tropics) are also regions with low cloud cover. It would be interesting to see if they agree at all with the observations.

    • ‘would cause a net increase of 3w/m2 in the energy budget’ –> should read decrease

    • Alberto. Thanks for your comment.

      On specific issues you raise:

      It’s surprising that cloud cover and temperatures would be positively rather than negatively correlated. iirc most people who look into the issue find the opposite

      Yes it is surprising, but I’m just going where the data take me. I don’t claim to understand what’s going on, but increased SSTs might be expected to increase cloud cover because of increased evaporation while the the clouds themselves act to reduce temperature. We therefore might have two opposing forcings acting simultaneously. This dovetails with the results of the second paper you link to:

      The conclusion is a frustrating ‘we cannot tell yet’: cloud feedback might be positive or negative depending on the time period and dataset used.

      It would also reduce your estimate of 3w/m2 cloud forcing to more manageable levels.

      I can’t do much with the first paper, which gives no cloud cover data and which covers only the period since 2000. The fact that an analysis based on only 15 years of cloud reflectance data “yields a most likely ECS estimate around 4.0 K; an ECS below 2.3 K becomes very unlikely (90% confidence)” also does not fill me with confidence.

  4. john PITMAN says:

    Perhaps the missing link here is ocean surface layer biogeochemistry – namely the link between dimethyl sulfide DMS and the formation of cloud condensation nuclei (CNN). Lovelock made this link in his original work on the gaia hypothesis, and more recently the following paper 9amoung others) has further explored this process:
    W. Richard Leaitch et al., Dimethyl sulfide control of the clean summertime Arctic aerosol and cloud

  5. john PITMAN says:

    One of the most comprehensive reviews of the DMS-CNN link is by Kloster

    and the paper by Lana et al provides additional mapping of oceanic DMS

    An updated climatology of surface dimethlysulfide concentrations and emission fluxes in the global ocean

  6. Grant says:

    At what altitude are we considering the effect of clouds?

    And which SST numbers is one comparing – the recorded set or the “adjusted” set (though it may not matter much for the comparison).

    At typical “sea level” altitude and a little above it cloud cover, assuming it is not associated with notably cold or warm air movements, cools compared to sunshine during the day but may warm (compared to a clear sky) at night.

    The adjustment effect, either way, probably varies in relative terms with changes in altitude even if all other factors are assumed to be constant.

    Given the claimed importance about knowing the answers I find it odd that no one seems to be suggesting setting up a more complete measuring system for global temperature than is currently available.

    After all it is not temperature that is being measured but a proxy for temperature form a satellite or two with some marginally reliable sensors that require significant understanding to interpret – leaving everything “open” to a closed debate.

    How difficult would it be to set up an array of geostationary satellites each measuring something simple for a few fixed locations and deriving a difference over time. After all that is all people are claiming to measure now. An amount of change.

    It doesn’t require many inputs to do that for an entire object. No complicated gridding and no complex questions to try to answer. Just a simple – “What is really happening?”

    Could be expensive? Well, yes but a few cancelled COP meetings while collecting a few years of data would likely pay for it in so many different ways.

    I’m somewhat puzzled that the UN, for example, has not suggested such a thing.

    Calculating the exact temperature of something as chaotic as the earth system is surely scientifically interesting for those involved but no absolutely necessary to know whether the general trend is up or down or neither.

    Currently the cost of only “knowing” the trend through the medium of modelling is probably far outweighing the cost of finding out. Why is it that that seems to be accepted by governments? Do they not want to know something more likely to be pertinent to their policy decisions?

  7. Stuart Ellison says:

    It’s fair to say that temperature and cloud formation is probably a two way street.

    You could say that either “controls” the other it just depends on your frame of reference.

    e.g. does pressure control volume or does volume control pressure? They are both variables in the same system.

    Clouds and temperature are just energy expressed either as a molecular vibration or a phase change.

  8. Javier says:

    Interesting Roger.

    Just thinking out loud, clouds are usually produced when warm humid air encounters colder conditions, so both warm temperatures for the water and colder temperatures for the air are important. The lack of correlation between 30-60°N is giving you the clue. Over land masses increase in temperatures reduces cloud amounts because the surface cedes the temperature to the air. Warm air accepts more humidity without making clouds. Over the oceans, it’s the sea who gets warm and keeps most of the heat, and thus clouds are made more efficiently. Over El Niño years, the entire El niño area is covered in clouds permanently and it probably rains a lot, and when those clouds get pushed over land it rains a lot there too.

    As surface creatures we tend to think mainly about the continental surface, but the climate of the Earth is decided on the oceans, and a warmer planet means a cloudier planet, with more precipitations (with some regional exceptions), not the opposite. And once the oceans are warm enough, CO2 plays almost no role, because there is so much water in the atmosphere that essentially saturates the absorption lines for the CO2.

    We know that the Eocene was not only warmer, but also a lot wetter. Why do we keep ignoring paleoclimatology?

    Latitudinal Gradients in Greenhouse Seawater d18O: Evidence from
    Eocene Sirenian Tooth Enamel

    Mark T. Clementz and Jacob O. Sewall

    The Eocene greenhouse climate state has been linked to a more vigorous hydrologic cycle at mid- and high latitudes; similar information on precipitation levels at low latitudes is, however, limited. Oxygen isotopic fluxes track moisture fluxes and, thus, the d18O values of ocean surface waters can provide insight into hydrologic cycle changes. The offset between tropical d18O values from sampled Eocene sirenian tooth enamel and modern surface waters is greater than the expected 1.0 per mil increase due to increased continental ice volume. This increased offset could result from suppression of surface-water d18O values by a tropical, annual moisture balance substantially wetter than that of today. Results from an atmospheric general circulation model support this interpretation and suggest that Eocene low latitudes were extremely wet.

    • Thanks Javier. Good commentary.

      Regarding your comment about the 30-60 “bust” being related to land masses, there is in fact a land cloud cover series – CRU TS3.23 – which can be accessed through KNMI Climate Explorer. I’ve looked at some of the raw records used to construct it and wouldn’t trust it very far, but globally it shows a gradual cloud cover increase between 1900 and 1980/90 similar to what ICOADS shows, but with a total increase of only about 1% instead of 10%.

      And the series for 30-60N shows roughly the same pattern:

      I’m not sure whether this supports or contradicts your theory. But if CRU TS3.23 is anywhere near correct cloud cover is now significantly higher over the oceans than over the land (~67% versus ~55%).

  9. stone100 says:

    Is it possible that clouds make the tropics cooler but make arctic and antarctic regions warmer?

  10. Pingback: Climate Partly Cloudy | Science Matters

Comments are closed.