Aboard a cargo ship steaming from Los Angeles to Honolulu, his radars spinning 70 feet above a dark sea, Ernie Lewis filled another latex balloon with a whoosh of helium and let it fly.
The balloon rose 15 miles, swelling in thin air, a fat white disc disappearing against the Milky Way. And then, as always, it popped.
An atmospheric scientist at Brookhaven National Lab, Lewis had joined the Spirit early in October for its regular shipment of 900 food-stuffed containers to Hawaii. And he had brought cargo of his own: three white shipping containers, given pride of place on the bridge deck, each carrying stimulus-funded tools aimed at a mysterious sky.
Lewis and his techs were testing every part of what will be a yearlong mission. For the first time, they will measure the sustained behavior of the lumpy pancake of clouds over the northeast Pacific Ocean. They had dozens of instruments to calibrate and balloons to launch every six hours. It led to an irregular schedule, but not a bad one.
"If it's inconvenient for us to get up at 3 in the morning, that's just too bad," Lewis said on the phone from Honolulu's Sand Island, halfway into his round-trip voyage aboard the Spirit. "It's what the science needs."
And -- more importantly -- it's what climate models demand.
For decades, scientists have known that clouds, and their response to humanity's carbon dioxide emissions, would be a roadblock to estimating the speed of global warming. Would feedbacks from higher temperatures cause more clouds, or fewer -- and where? Even without considering human pollution, it's been uncertain. Computers lacked horsepower to model turbulence that drives clouds beyond a small scale, and so climate models have done their best to approximate these nuances at the global level.
It hasn't always gone well.
Time and again, clouds have provided the most intractable disagreement among models. Over simulated decades, sometimes the clouds accelerated warming; sometimes they were unchanged; and rarely, they dampened the heat. But most of all, they told their creators that they had failed to solve the problem in a convincing way. And the biggest point of disagreement has been low clouds like those off California's coast.
"This really is the fundamental problem in climate models," said Joao Teixeira, deputy director of climate science at NASA's Jet Propulsion Lab and a collaborator on Lewis' Spirit project. "How do you represent the clouds?"
There are ways the models do agree. All project that the wispy ice clouds in the upper atmosphere will rise in height, causing more warming, a result seen not just in silicon, but in the real world. But without an accurate handle on low clouds, Teixeira added, it's impossible to fix on the planet's exact climate sensitivity, a shorthand used by scientists to test Earth's warming under instantly doubled CO2 emissions.
"It's a completely unsolved problem," he said. "It could easily be by 4 degrees [Celsius], or 2."
Even the primary mechanism of this cloud feedback is in doubt: The change could come from higher surface temperatures, which increase in a laggy, variable way due to the ocean's heat retention and natural variability, or it could be governed by fast feedbacks, CO2 emissions changing the clouds even before they warm the Earth. Even more likely, both dynamics play together in an irreducibly complex double Dutch.
Still, cloud scientists are game. Using historical records, ground radar, satellite data and advanced supercomputers, these researchers are improving the models and probing for any sign of past change in the clouds. And although this work has not led to a decreased spread in the climate sensitivity, which will remain about the same in next year's U.N. climate science report, it has been far from for naught.
"The uncertainty in cloud feedbacks hasn't gone down. It's sort of stayed the same," said Chris Bretherton, an atmospheric scientist at the University of Washington. But that's not the same as stalled progress, he added. "We're taking what was before uncertainty that lay outside of the models, and putting it into the models."
These questions pivot especially on marine stratocumulus clouds, sheets of bulbous gray gruel that rest low above a fifth of the Earth's surface, reflecting the sun. And there are few better examples of those clouds than the transect between Los Angeles and Hawaii, where a remote, eerily stable stretch of stratocumuli slowly dissolves, on approach to coconut trees, into cotton-puff cumulus and open sky.
For decades, scientists have known they'd need to take their cloud tools off the shore and into the ocean, but until last month, those efforts had stalled. Instead, work funded by the Energy Department has been earthbound, based in Oklahoma, or at best on remote islands, where clouds remain shaped by even the dullest contours of the land.
"There are very few measurements over the ocean for any period of time," Lewis said.
There is an enormous pent-up demand for the data Lewis and his team will find along the Spirit's path. The project, nicknamed MAGIC, will provide evidence for teams of climate scientists and modelers waiting to test their theories against reality. A successful, stable voyage is crucial. It is one of the best hopes left to help resolve the cloud enigma.
So far, scientists have found no other shortcuts to solve the problem.
But, of course, that doesn't mean they've stopped trying to find one.
'A lot of noise'
A couple of years ago, at a sleepy hotel outside Norfolk, Va., Andrew Dessler, a climate scientist and data wonk at Texas A&M University, was absently reading Watts Up With That, a climate contrarian website, as he did from time to time.
On this night, though, the site didn't just rile him up. It provided inspiration.
As part of its antagonism to climate science, one of the site's writers singled out the world's sharp temperature drop in January 2008. Dessler knew the fall was largely due to natural variation in the Pacific. But one contrarian's evidence, it turned out, was another scientist's eureka. This was an opportunity, Dessler realized. He had wanted to study global cloud feedbacks but lacked the recent warming needed to do the work.
"For years I had been waiting for a big El Niño," Dessler said in a recent interview. "Suddenly, I realized that a big drop could work as well as a big rise."
From year to year, much of the planet's temperature shifts are controlled by the El Niño-Southern Oscillation, a somewhat predictable pattern of circulations in the Pacific Ocean. During El Niño, temperatures run high, while La Niña, its sister phase, cools things down. It's a dynamic Dessler had previously used to study how much more water vapor increased emissions will allow in the atmosphere, causing additional warming.
From that moment of clarity in his hotel, Dessler eventually published a study that attempted to gauge, with a decade of light and heat observing records from NASA's Terra satellite, whether the Pacific could do for clouds what it had done for water vapor.
And, he found, it could. Well, kind of.
Published to much notice in Science in late 2010, Dessler's study found that the cloud feedback would likely make warming worse -- or, more importantly, he found it unlikely that cloud changes would push temperatures down. Without such a response, the clouds could not cancel the heat stemming from increased greenhouse gases. The world, resilient as it is, would not make up for this particular human insult.
The error bars attached to Dessler's estimate were large: His results pointed positive, but it still was possible for clouds to have a slight negative feedback, too. And he was quick to admit a deeper problem: El Niño may be a poor proxy for long-term warming. After all, those same circulation shifts in the Pacific could change clouds in a way that the steady accumulation of carbon emissions would not.
Since his big news splash, Dessler has been taking deeper dives into the data. And what he's found has exposed the limits of short-term satellite studies and called his past results at least somewhat into doubt.
Using precomputed cloud tables, Dessler has a study submitted to the Journal of Climate that parses the climate response of various cloud types over a decade of satellite data. It's rigorous work but mostly shows a decade can't say much about clouds.
Unlike his 2010 paper, the new study finds an overall cloud response that is slightly cooling, but statistically indistinguishable from zero. (For the climate-fluent, that's a net feedback of -0.12 W/m2/K, with an error of 0.78.) And as most models point out, the response is dominated by those clouds hovering over the Spirit's path.
"It really is low clouds that are driving the cloud feedback over the last decade," he said.
Dessler believes the net zero result is due to a bias in the cloud tables more than anything else. His 2010 study remains the better estimate of the feedback, he said. But the work goes to show just how tricky it is to get at clouds from only satellites.
It's a 1 percent problem. Compared with greenhouse gases, clouds reflect and absorb a huge amount of solar energy. Small statistical shifts loom large. That can make it seem as if clouds aren't tied to global temperatures -- even if they are, as his new study shows.
"The scatter comes from the fact that the net impact of clouds is a small difference of two really big numbers," Dessler said. It's easy for random events to skew it all. "If you change any of those big numbers by a very small amount, you get a lot of noise."
Other scientists have also remained skeptical of El Niño's usefulness as a proxy. The University of California, Los Angeles' Troy Masters, for one, published a study critiquing Dessler's 2010 paper. And scientists more deeply involved in the nitty-gritty of cloud theory find it doubtful that Dessler's methods could ever give any real certainty on the feedback problem.
Dessler himself admits that the way he's attacked the problem, it may be impossible to refine his estimates further. But his work is enough to rule out the extreme low estimate that the world will warm only 1 degree Celsius under doubled emissions, he said. The notion of clouds as warming's savior is out the window.
"You can exclude with high confidence some of the claims made by skeptics," he said.
But to get a handle on past cloud change, a much longer record is needed.
Fortunately, one exists.
'Not a lot of agreement'
There's nothing glamorous about the daily grind of weather station work.
Each day, trained observers stare into the sky and write down what they find. From Siberia to Africa, the work is similar, following standards that have remained nearly identical for decades: Determine the percentage of cover and then flag one of nine cloud types at three altitudes. It's an anecdote of the sky. And since the 1950s, there have been almost 300 million anecdotes.
For the past three decades, a group of scientists have tried to harness the potential power of this human record to search for changes in cloud cover. Based out of the University of Washington, they have painstakingly assembled what amounts to a cloud atlas: a record of observations made from 5,388 weather stations since 1971, and ships since 1954.
The work isn't easy. Humans are far from perfect, and spurious trends appear all the time, said Ryan Eastman, a staff scientist at UW. Eastman has spent much of his early career, since he was an undergrad, taming this mongrel record.
The best example of bad data comes from North America. A line of stations in the Arctic boundary appeared suddenly shrouded in clouds, a bunch of San Franciscos sprouting among the polar bears, their cover growing by 20 percent in one decade. It couldn't be real, Eastman thought. And it wasn't. Every station showing a cloud spike was part of a Cold War system scouting for Soviet nuclear bombers. Someone decided the stations should only record stormy weather, and that was that. Human error strikes again.
"It was a procedural change that drove a fake trend," Eastman said.
All this quality control done -- Eastman killed off similar quirks in Russia, though their cause remains mysteriously Cyrillic -- the UW group does have a modest result, reported this summer: Over the past four decades, the world's total cloud cover has declined about 0.4 percent a decade.
It's not a huge trend. And that's part of the point.
"There hasn't been a huge change in cloud cover over the last 40 to 50 years," Eastman said. "We're kind of the null hypothesis. Satellite records have shown a bigger trend."
Here it gets confusing. Satellites have shown a bigger change in the opposite direction, typically finding more clouds forming over the eastern oceans today than in the mid-1980s. The two records fundamentally disagree. The atlas supports fewer clouds forming under warming, and the satellites support more. Or they could both be wrong.
"There's still something weird about the satellite and surface record," said Bretherton, the UW climate modeler. "They don't naturally agree with each other at all. And so it makes us worry about both of them."
Composed of several different eras of instruments, satellite records are by nature fickle creatures. There's an art to cleaning them up and stitching them together, just like the atlas. It's a talent held by Joel Norris, a climate scientist at the Scripps Institution of Oceanography, who is regularly pestered to publish more about his arcane methods as he tries to put together a calibrated record longer than a decade.
"There's not a lot of agreement on what the atmospheric changes have been," he said.
A better satellite record could solve those disagreements, but it's not easy to do.
Norris rattled off some of the common problems.
Take one long-term record, he said. It's based off a changing number of stationary satellites. The more there are in orbit, the more each looks straight down -- and it's harder to see clouds at a dead-on angle. Fix that. Other satellites are notorious drifters: What's seen at noon one year appears at 2:30 p.m. in a couple of years. Because clouds wax and wane daily, that rhythm can be mistaken for a trend. Tweak it. And some satellites will, inexplicably, see a huge uptick in clouds everywhere. But it's not nuclear winter or a super-volcano eruption. It's just a calibration error.
There is one area where the weather stations and satellites may be starting to agree. Eastman's study found that over the past 40 years, the jet stream has nudged toward the poles, expanding the tropics and pushing storm tracks northward, a result predicted by climate models. The intrusion of the subtropics into the midlatitudes could be driving, in turn, the drop in cloud cover.
"Our changes are significant, but small," Eastman said.
Those results are echoed by a satellite study being prepared by researchers at Lawrence Livermore National Lab, who found that over the past 30 years, there has been a decrease in cloudiness between the 30th and 40th parallels -- in the north, the latitudes from Florida to New England. There's debate whether this expansion is due to global warming or changes in ozone, but the former is likely a player.
Still, there are limits to how far climate scientists will go in trusting either report. They suffer from the same flaw: They are weather systems pressed into double duty as climate monitors. As an engineer would say, they are a "kludge."
Not only that, for the atlas, it's kludge that may not last.
Eastman isn't quite sure how the cloud atlas will survive when its guiding light -- Stephen Warren, who began the project decades ago -- moves toward retirement. Certainly, it will have to get more automated. But automation may also kill it. When the United States moved to computerized weather stations, they had to kick that less descriptive, laser-based data out of the atlas. It's a trend that will only spread.
Maybe if a rigorous climate observing system appears -- something the world does not have right now -- and runs for another 30 years, while global warming continues, it might begin to tell a clear story about the cloud response, said Ray Pierrehumbert, a climate scientist at the University of Chicago.
"But it's just not a viable option to wait 30 years," he said.
And if scientists can't look to the past, or wait for the future, it's up to the present.
That means, right now, it's up to an aging cargo ship en route to Hawaii.
Aboard the Spirit, a day out of Los Angeles, the clouds did not disappoint.
Strung between the turbulent boundary layer and the free air above, long stretches of marine stratocumuli coated the sky, hanging above Lewis' bald head. But not as many as he expected, and far from a uniform sheet: The ship sailed under closed cells to open blue, and back again. It was so variable, in fact, that Lewis planned to shoot photos of the sky every 10 minutes or so on the journey back, to show his fellow investigators.
Just looking at the lumpy stratocumuli begins to reveal their nature. There's more to them than formless stratus, that glorified, elevated fog. Stratocumuli are driven by the weak currents of convection, pinned at the boundary of the ocean's turbulent influence by stable dry air and drastic cooling of its own creation, due to its reflectivity. Viewed from the surface, each cell runs the size of a fist extended toward the sky.
There are no semi-stable sheets like these over land, which is why government scientists had long desired to put their mobile lab at sea. Once stimulus funds paid for a second clutch of containers explicitly designed for ocean work, all they needed was a ship. And the 32-year-old Spirit, operated by Horizon Lines and near its final appointment with a dry dock, was the perfect candidate. Unlike modern vessels, which leave no open deck space not flagged for commerce, the Spirit had a generous bridge deck, perfect for science.
It also provides quite a view.
"I tried to get out there by 6 to watch the sunrise," Lewis said.
The trip wasn't all cloud watching, though. There were problems to fix.
Three scientific containers, lined with railings on top to hold radar and lasers, were a snug fit on the bridge deck. A tool chest latch snapped one day, spraying drawers on the ship's rolls. A pump broke. Some instruments, their recordings corrected by an 8-square-inch box measuring the ship's jerks, proved recalcitrant. And that wasn't even including the microwave radar mounted on a hydraulically balanced table meant to keep it level no matter the waves' strength. The stable table would wait till later.
Then, there were the weather balloons, needed to describe the wind, heat and humidity of the air above. Original plans had called for launching them from the ship's stern, some 800 feet behind the bridge, but logistics had restricted them to the bow, where high winds threatened to bang the balloons into the big bank of containers, stuffed with perishables and cattle feed, trailing behind. But they got the technique down.
"When we had a first balloon launch that worked," Lewis said, "it was very exciting."
It was an easy cruise, Lewis added. Over calm waters, they saw few ships, or signs of life -- a couple of dolphins, some seabirds. The ship's 23-member crew was curious, the captain supportive and the meals hearty: flank steak one night, Cornish game hen the next.
"Our biggest problems are which entrees to get for dinner," he said.
Soon enough, after a couple of days of balloon launches and sunrises, they came to the point of mystery. Puffy cumuli began to form under the stratocumulus pile, then the closed fists started to disperse, like a protest crowd in a police state. The cloud pattern migrated from closed to open, letting the dark ocean absorb the sun's rays. By the time the Spirit had reached Hawaii, only rare trade wind cotton balls floated in the sky.
No one understands why this transition happens or how it could change in the future.
"This may be the key question for the cloud-climate feedback," said JPL's Teixeira. Under warming, it could well be that this zone of change could move closer to Hawaii, or closer to California, he said. "Trying to understand this transition is really important."
There are a few ways of getting at it, Teixeira said. First, the microphysics: What are those puffballs doing to the stratocumulus sheet above? They are feeding water upward, but also drawing air down from above the cloud tops; one could grow the clouds, the other could kill them. Which is it? And then there's the climate angle. Do the marine stratocumuli simply form in regions with cold water, and dry air above? And if either is true, how does it change under increased greenhouse gases?
For this reason, some of the most important data captured by MAGIC will not solely be its records of the clouds. It will be the habitat they live in. Like an ecologist measuring a forest's humidity and soil fertility, cloud scientists need a statistical baseline for the Los Angeles-Hawaii transect: sea temperatures, surface water fluxes, atmospheric humidity -- anything to help define the atmosphere's state.
Scientists, above all, need data. Lewis and company will feed them soon.
Then, perhaps, they can start being certain their models aren't always wrong.
If only the world had infinite computing power, there wouldn't be a cloud problem.
When it comes to clouds, there are two types of computer models, and it's easy to get them mixed up. There are the global climate models familiar to most, which chop the Earth into 100-square-kilometer grids, simulating the planet for years on end. Much less known, however, are the small-scale cloud models that divide the atmosphere down into boxes of 10 meters square, where the computer can begin to capture the chaotic atmospheric turbulence that rules the clouds.
Those cloud models are so demanding, however, that no computer can run them on anywhere close to a global scale. It's a dream, but not one to be realized soon.
"If I could run a [cloud-resolving] model globally, or even over a huge area, I might be able to get a result I at least didn't know was wrong," said Joyce Penner, a longtime modeling expert at the University of Michigan. "But we can't do that yet."
That's not to say the climate models get it all wrong.
Over at Lawrence Livermore National Lab, Stephen Klein, a research scientist, puts the Energy Department's brute computing power to good use, operating a test bed where he takes climate models and runs them like weather forecasts, probing their rendering of clouds. Except for the time involved, the two models are similar; most errors in climate models become visible after two to three days of simulation.
Klein recently finished putting all the models going into the next U.N. climate science report through their paces. They disagreed on the low clouds, of course, but they were unanimous in predicting that the thin ice clouds at the atmosphere's top, which act like greenhouse gases, will rise in response to warming, causing a positive warming feedback.
"That's something all the models robustly predict," Klein said. Indeed, he added, "the high clouds rising is ... the thing we're most confident about climate change [and clouds]."
(At first blush, that may seem to run against a study published earlier this year, to some media notice, that found high clouds to be falling in height. But Norris, the savant of satellite correction, took a close look at that data and disagreed. NASA's Terra satellite had known syncing problems with its cameras that caused it to ignore low elevation points. Correcting for that bias revealed that cloud tops were in fact rising.)
Klein also compared these climate models with their decade-old code, using his test bed to get at how well each re-created observed clouds from the satellite record. For a long time, models have been notorious for having sparse clouds, with the ones they did create reflecting far more light than they do in reality. It's known as the "too few, too bright" bias. And it's a problem that's quietly slipping away.
"Clouds are not as bright as they used to be, and they're getting more of them," Klein said.
Indeed, his work has hopeful news for modelers: They're getting better.
"This is clear evidence that work into clouds has gotten better over time," he said. They still don't agree on all the feedbacks, but it's progress. "Even if we're not fully reducing the uncertainty, we're certainly making the clouds we have today better."
At this point, Klein can only speculate why they've improved. He doesn't write the code. He only tests it. Maybe it's a better representation of the clouds' microphysics, the fine-scale process of how vapor becomes a cloud, and back again. The models have also refined their vertical resolution -- perhaps fewer clouds are rendered 500 meters thick, when they're really only 100 meters tall. And it could be they've better represented how to get at the uncertainty inherent in grids often half covered by clouds.
Despite these improvements, and the vast political pressure created by the U.N. reports to increase their accuracy every five years, the models, including those tested by Klein, have far to go. They don't even reflect much current science, Teixeira said.
"While we know much more than we did, we haven't been able to implement that into models in a way that's satisfactory," he said. "That's as much an engineering problem."
Teixeira talked with his peers about the problem at a recent workshop, and there were no obvious solutions. Around the world, there are maybe 150 people working on serious climate modeling, to be generous, with 20 or 30 of those doing cloud work. Teixeira compares that with one of JPL's other projects, the most recent Mars rover, which took hundreds of engineers to create, none of whom was under pressure to publish or perish.
"These people are not spending their time trying to write a paper," he said.
Would a similar corps of engineers help the cloud watch? Maybe. Maybe not.
All is not lost, though. There could be another way of improving the models, one that can finally stitch together the short-term measures of those clouds over the Spirit with the silicon future of supercomputers. It's a technique that goes to the roots of science.
To understand the marine clouds, scientists will first have to tear them apart.
Understanding why there's so much uncertainty
Until two years ago, Bretherton had no hope of doing the work he does now.
The University of Washington modeler has long used intense mathematical models called large-eddy simulations to get at the chaotic fluid movement of the atmosphere. There's so much math, however, that no supercomputer could run a detailed version long enough to get at the cloud problem -- until a couple of years ago. Now he can run an eddy model out to 10 days, long enough to make things interesting.
They're not re-creating all the world's turbulence. They don't need to.
"We're just trying to simulate a postage-stamp region off the coast of California," he said.
And comparing these models, Bretherton can, at least, see the end of the beginning.
"One of the most important things we learned from doing this work," he said, "is that I think we understand the reasons there's a lot of uncertainty. ... Basically, there are processes that work both ways -- ones increasing clouds and ones decreasing clouds."
For instance, in a warmer climate, the middle and upper atmosphere heat more than the surface. That will strengthen the "lid" on marine clouds, making mixing harder and increasing cloud cover. On the other hand, increased CO2, even before it warms the surface, will trap more heat in the clouds, weakening the process that drives their formation. Any battle like that makes it difficult for models to get it right, though notably all the simulations agreed that there would be no negative feedback.
If that's all Bretherton did, it'd be notable. But it goes further. It goes back to MAGIC.
Rather than designing a model and letting it run to see what happens, Bretherton and his peers -- all part of the massive CIGLS project -- took advantage of their direct physics to change the habitat of those marine stratocumulus clouds. One run, they tweaked sea temperature; another, humidity; or subsidence; or wind speed. Carving each variable out could, at last, connect the work back to the real world, Bretherton said.
"If we break the problem down, it might be easier to test in the present day," he said.
And, as it happens, out in the Pacific, such measurements are happening right now.
Another old hand with turbulence math, Teixeira plans to use all those statistics Lewis and his technicians on the Spirit are collecting to compare how the clouds changed during the seasons with what the eddy models show. And what they learn, they can then build back into the global climate models.
A native of Portugal, a culture well-known for its naval exploration, Teixeira plans to join the Spirit for one of its Hawaii voyages himself. He and Lewis want to have modelers and theorists on board as often as possible, so they can understand how MAGIC made its measurements. In a 1 percent problem, every detail will be important.
And until then, give models a break, Teixeira added.
"They're not perfect," he said. "But they are the only solution to the problem."
Next: Cloud chambers, cirrus and a crystal method
Second in a series; click here for Part 1.