A Detective Story, Chapter 2: The Questions That Nobody Asked

Satellite-Era Mismatches in Atmospheric Power Assessments

Confirmation Bias

We pick up our story from the moment when Edward Lorenz used the most plausible assumptions about the distribution of heating in the atmosphere to estimate the maximum amount of work per unit time (aka power) that this heating could sustain in Earth’s atmosphere. He obtained an estimate that was about half the power independently estimated from the observed winds.

Again, the maximum possible power was already about half the observational estimate.

Before we make a leap of half a century to consider more detailed estimates, I would like to share some thoughts about confirmation bias. When Lorenz set himself the task of calculating the maximum possible power from differential heating, he used the best available data and the most plausible assumptions. Had his simple model yielded the right magnitude for maximum power—for example, 0.03 of solar power versus the observed 0.02—there would have been little reason to search for other assumptions or finer resolution.

Since data are rarely perfect and usually come with uncertainty, how confirmation bias can interfere is as follows. A researcher uses a plausible set of parameter values to test a hypothesis. If the result matches the observations—hurrah, we can publish it. If it does not match, the researcher may look for another set of values for the same parameters, still within the uncertainty ranges and still plausible, although not the first choice. Then, for a certain choice—voilà—the hypothesis matches the observations. This can also be published, but the fact that the first plausible choice did not work may not always be mentioned.

This type of confirmation bias is not very visible, but it can matter. It is relatively innocent in the sense that, if the ultimately chosen set of parameters remains within observational uncertainty, the hypothesis is at least capable of explaining reality. Until it is shown to conflict with other evidence or with physical laws, it remains one possible explanation. However, as uncertainties tighten, the inability of the most plausible parameter values to explain the data may hint at where the error lies.

The fact that Lorenz reported the very low value produced by his first choice suggests that he might have regarded the calculation as robust enough to deserve attention, despite the discrepancy.

Eddies Eating Up Power

Lorenz attributed the discrepancy to assumptions that were too simple:

“Ignoring variations with height and with longitude, and assuming a reasonable value for the mean albedo, we find that the maximum possible rate of generation of available potential energy in such an atmosphere is about 0.01 of the incoming solar radiation.
This is less than the estimated value of the actual rate of conversion in the real atmosphere. If the estimate for the real atmosphere is correct, it follows that we have failed to take some factor into account.”

Half a century later, satellite observations allowed the same question to be examined using more detailed estimates of atmospheric heating. In 2013, Romanski and Rossow published their analysis:

“Our estimates of G are the only ones that are computed from diabatic heating calculated directly from observations and the only ones that decompose G_z and G_e into the contributions from each separate heating process, thus permitting evaluation of the role of cloud processes. Since we calculate diabatic heating from observations, we can link the various contributions to the physical processes that generate them and to the atmospheric circulation, so that we can diagnose, in a physically meaningful way, how the atmosphere’s circulation modifies the spatial distribution of heating away from the simple pattern of net solar radiation.”

Here by G the authors mean generation of available potential energy by differential heating, that is, the vertical red arrow in the schematic picture that we discussed in the previous chapter. G_z is the zonal-mean generation, associated with the meridional distribution of heating and temperature. G_e is the eddy generation, arising from departures from the zonal mean, including the midlatitude eddies.

Uneven heating generates available potential energy by “lifting the atmosphere up.” Kinetic energy is then generated as air flows down along the resulting pressure gradients.

That is to say, in his maximum-power estimate, Lorenz considered only the meridional differences in heating. But at each latitude there are also variations with longitude—in temperature, heat transport, and absorbed and emitted radiation—associated with atmospheric eddies.

Now, in our detective plot, resolving these longitudinal variations could in principle raise or lower the estimate. Lorenz had in fact suspected that the eddy contribution might be negative, but its magnitude could not be reliably established with the data available at the time.

Romanski and Rossow were able to calculate whether the longitudinal heating patterns associated with eddies “lift” the atmosphere further, creating more available potential energy, or instead reduce this energy by preferentially heating colder anomalies and cooling warmer ones.

To keep some suspense before announcing what Romanski and Rossow actually found, let us discuss those eddies. What are they?

If we look at the longitudinally averaged global circulation, we can see three circulation cells in each hemisphere. We already know that warmer air is “higher” than colder air, so the “highest” air resides in the tropics. In the Hadley and Polar cells, the air aloft moves “in the right direction”: from warmer to colder regions, down the pressure gradient.

A very schematic depiction of the zonally averaged atmospheric circulation.

But in the Ferrel cells, squeezed between the “correct” Hadley and Polar cells, the motion is in the opposite direction: from cold to warm—that is, uphill! Kinetic energy is naturally consumed in this uphill motion.

The images below show the instantaneous winds and cloud patterns from which the zonally averaged circulation emerges. The eddies are particularly evident in the Ferrel Cell in the Southern Hemisphere.

The visualizations are from the Earth Nullschool project for July 31, 2026

In the second graph, the white shading shows total cloud water.

The Questions That Nobody Asked

Romanski and Rossow found the zonal-mean generation to be¹

G_z = 1.5 W m⁻².

This was already considerably below Lorenz’s estimate of 3 W m⁻². Then came the eddies. Instead of adding the missing power, they contributed

G_e = −0.3 W m⁻².

Thus, the total generation of available potential energy by differential heating was

G = G_z + G_e = 1.2 W m⁻².

The eddies did indeed eat up some power. What does this mean?

Positive G_e is produced when heating strengthens the longitudinal temperature differences: warmer-than-average regions are heated, or colder-than-average regions are cooled. Negative G_e means the opposite: warmer regions are cooled or colder regions are heated, so that the temperature differences are reduced.

This is not the same as the negative kinetic-energy generation in the Ferrel cells discussed above. There, air moves against the pressure gradient and kinetic energy is spent. Here, it is the distribution of heating and cooling that destroys eddy available potential energy.

While these are two different ways in which eddies can be said to eat up power, they are not independent and may in fact help explain why the eddy generation is negative.

Indeed, as we discussed earlier, for a warm location to receive net heating, it must absorb more energy than it emits. In a steady state, the excess heat has to be transported away to a colder location. This is what happens in a heat engine: heat flows from warm to cold, and part of this flow can be converted into work. In the atmosphere, this is the case for the thermally direct Hadley and Polar cells.

There is, however, an important subtlety. When air carries heat away from a warm location, that location, in the steady-state, is left with a positive local heat budget: absorbed solar radiation exceeds longwave emission. At the colder location, the imported heat must be lost by radiation, so the local heat budget is negative. Thus, the circulation does not merely respond to differential heating. By transporting heat, it also helps create the very pattern of differential heating from which the generation of available potential energy is then calculated.

The Ferrel cells work in the opposite way. They do not behave as heat engines, but as heat pumps.² A heat pump consumes work to transport heat from cold to warm. This is how the fridge in the kitchen works: electric power is used to remove heat from the cold interior and release it into the warmer room.

If most of the eddies contributing to G_e​ are associated with the Ferrel cells, their heat transport should tend to produce the opposite heating pattern: net heating in colder regions and net cooling in warmer regions. This gives negative generation of available potential energy.

After half a century, the neglected longitudinal structure had finally been resolved. It did not repair Lorenz’s low estimate but made it lower.

“If differential heating generates only about 1.2 W m⁻² of available potential energy, how can it sustain the considerably larger power independently diagnosed from the observed winds and pressure gradients?

Is an important heating process missing, are the dynamical estimates too high, or is the premise that all atmospheric work is generated by differential heating incomplete?”

These are the questions that nobody asked.

Outlook

When thinking about the subtitle for this post, I vaguely recalled Agatha Christie’s novel Why Didn’t They Ask Evans? Hence, The Questions That Nobody Asked. But this is not a direct analogy. What seems important here is that there has been almost no discussion of the published mismatch between the prevailing thinking and the data. This is an informative illustration of how our concepts shape what we choose to see and what we choose not to see.

There are many publications evaluating the Lorenz energy cycle. The most recent one I located was published in 2026 and compares atmospheric power across four major atmospheric datasets. Romanski and Rossow are included in the reference list, but the approximately twofold discrepancy is not discussed.

Romanski and Rossow estimated the generation of available potential energy by differential heating at about 1.2 W m⁻², less than half the wind power independently estimated from observed wind speeds and pressure gradients. Thus, the mismatch first encountered by Lorenz in the late 1950s appeared again in the work of Romanski and Rossow half a century later. They may not have recognized that Lorenz had faced essentially the same problem; at least, they did not cite his 1960 work.

Ilya Glazunov. Metro

In retrospect, the progress of ideas is often depicted as a bright individual sacrificing comfort for the sake of truth while confronting an aggressive majority. A visual analogy that came to my mind, also imperfect, is Glazunov’s painting above, with its contrast between the massive dark flow moving downward and the small light movement upward.

But I feel that the actual situation can often be considerably more peaceful and prosaic. Evidence that does not fit may simply fail to attract interest. After all, scientists are free to work on the questions they choose. No individual scientist bears personal responsibility for a mismatch between the prevailing paradigm—all atmospheric work is due to heating—and the data. Many therefore choose to do other things, as though the mismatch did not exist. Eventually, almost nobody knows that it exists and that something should be done about that³.

Then, when a new concept appears—in our case, the proposition that atmospheric motion is driven by condensation—the mainstream may appear as a monolith, although it is not one. From the beginning, the comparison is therefore not entirely fair. The mainstream is presented as though it had no inconsistencies, and authoritative figures may criticize the new concept from that position. Again, this need not result from anyone’s malicious intention. It may simply be how human thought and scientific attention work.

How, then, do things change?

I am struggling to keep the interest of readers who normally come here to read about how green life works its wonders rather than about abstract meteorological concepts. But for me there is no border between science and real life. This is one of the lessons I learnt from Victor Gorshkov. Science is an intellectual window into the beauty of Life. I feel this strongly and try to communicate it here.

Once one sees this beauty, it becomes impossible to unsee it, to forget that when it rains, the atmosphere does its work, and to return to the convenient picture in which heating alone does everything and the vegetation cover is reduced to a surface with a given roughness and color. That’s why we don’t stop.

1. The authors discussed several sources of uncertainty and performed sensitivity tests, but did not provide uncertainty estimates for the reported values of G_e, G_z, or G.

2. For scientific discussions of atmospheric heat engines vs heat pumps, see Huang and McElroy (2014) and Makarieva et al. (2017).

3. More recently, Zhang and Rossow (2023) noted that the new global radiative-flux profile data set could be used to assess the general energetics of atmospheric circulation “updating Romanski & Rossow, 2013”.

Related reading: