WorksPhysics and Astronomy
The Alpha–Omega Problem in the Assessments of Atmospheric Power
Abstract
Atmospheric power is commonly assessed with two Lorenz-cycle formulations treated as equivalent: the formulation, yielding gas expansion power , and the formulation, yielding kinetic-energy generation by horizontal pressure gradients, . We show that, for a statistically stationary global moist atmosphere, , where is the power of lifting precipitating water and is power associated with departures from gas-phase hydrostatic balance. Since W m globally, equality would require , so the associated vertical force imbalance would have to do negative work globally. Using six-hourly ERA5, JRA-3Q, and MERRA-2 fields for 2025, we find that resolved varies by a factor of 1.7 (1.9–3.2 W m), versus 1.2 for (2.4–2.8 W m). Zonal-mean profiles have area-weighted contributions near W m, making global a cancellation residual, whereas profiles vary much less with latitude. The difference ranges from to W m. Thus reanalysis-derived cannot provide an independent constraint on atmospheric power. Two treatments of the omitted surface layer give cross-reanalysis mean of 2.8 and 3.1 W m, with corrected values spanning 2.6–3.5 W m. Both exceed the observation-based Lorenz available-potential-energy generation estimate of 1.2 W m. Assuming , adding W m gives total atmospheric power of about 3.8–4.1 W m, motivating investigation of additional power-generation mechanisms.
Introduction
The rate at which the atmosphere generates mechanical power is a basic measure of its circulation intensity and provides a quantitative constraint on the mechanisms that maintain it. Published estimates of atmospheric power, however, differ substantially in magnitude and do not always refer to the same physical quantity.
Most quantitative estimates have been made within the Lorenz energy cycle (LEC) [16, 24, 1, 20, 11, 25] (Li, Ingersoll, Jiang, Feldman, & Yung 2007; Ma, Lembo, & Franzke 2021; Oh et al. 2026). In the Lorenz framework, differential diabatic heating generates available potential energy (APE), which is subsequently converted into kinetic energy by atmospheric motion. For a statistically stationary global atmosphere, the total generation of APE must balance its net conversion to kinetic energy. In the classical four-reservoir cycle, this net conversion is
Here is the global mean APE-to-kinetic-energy conversion rate per unit surface area; and are the zonal-mean APE and kinetic-energy reservoirs, respectively; and are the corresponding eddy APE and kinetic-energy reservoirs; and denotes the conversion rate from reservoir to reservoir . The corresponding APE generation is , where and denote diabatic generation of zonal-mean and eddy APE, respectively. In statistical steady state the classical LEC requires .
The conversion side of the LEC, , has commonly been evaluated using two formulations. The first is the pressure-velocity form,
where is Earth’s total surface area, is atmospheric volume, is air pressure, and is the material derivative. For the statistically stationary global means considered here, with negligible pressure-work flux through the control boundary, this quantity represents gaseous expansion power. Under hydrostatic balance it can be written on pressure levels as
where is the specific volume of gaseous air, is its density, is gravitational acceleration, denotes Earth’s surface, and and are the top and surface pressures. (3) is the alpha–omega form commonly used in LEC calculations.
The second formulation gives the rate of kinetic-energy generation by horizontal pressure gradients,
where is horizontal air velocity and is the pressure gradient. Transforming the volume integral to pressure coordinates gives
where denotes differentiation with respect to horizontal position at fixed pressure and is geopotential. Unlike (3), this transformation does not require hydrostatic balance. (5) is commonly called the formulation in the LEC literature, where in this conventional label denotes horizontal velocity; here we denote horizontal velocity by .
The generation and conversion sides of the cycle can in principle be estimated independently. APE generation can be diagnosed from atmospheric temperature and diabatic heating rates, the latter being constrained by radiative, sensible, and latent heat-flux fields. The conversion side is constrained by wind, pressure, geopotential, and temperature fields. Because global fields of diabatic heating are much less directly constrained observationally, most global LEC estimates have used the conversion side of the budget [29]. Reanalysis-based estimates of from these pressure-level formulations are generally around 2–3 W m, with appreciable dependence on dataset and methodology [20, 11] (Li et al. 2007; Ma et al. 2021; Oh et al. 2026). Outside the LEC framework, MERRA-based estimates include kinetic-energy generation by horizontal pressure gradients, , of about 2.5 W m [7], and a moist-atmospheric heat-engine work output of about 3.7 W m [14]. In contrast, the observation-based APE-generation components reported by Romanski and Rossow [29] sum to only W m. Whether these differences reflect distinct physical components of atmospheric power or differences among the diagnostics is therefore a basic issue for interpreting the atmospheric energy budget.
For a statistically stationary dry hydrostatic atmosphere obeying source-free continuity, complete global integration with consistent boundaries gives [16, 28]. The two expressions have therefore generally been treated as equivalent formulations of the same Lorenz conversion. Nevertheless, discrepancies between their empirical estimates, of varying magnitude, have been noted and discussed in several contexts when analyzing pressure-level reanalysis data. Li et al. (2007) identified reanalysis pressure vertical velocity as having relatively low data quality and, to take advantage of the higher data quality of horizontal velocity, transformed and into horizontal-wind form using the continuity equation. Marques et al. [20] found the two formulations to be comparable only when below-ground pressure-level data were excluded from the calculations. Y.-H. Kim and Kim [11] obtained broadly comparable global conversions using the alpha–omega and horizontal-wind formulations, although differences remained among individual conversion terms.
Thus, these discrepancies have generally been interpreted as arising from data quality or implementation rather than from a failure of the underlying equivalence. Using an ECMWF forecast model, Steinheimer et al. [30] characterized the two formulations as hydrostatically equivalent, compared them in selected cases, and found their global averages identical within error margins, although without reporting this analysis in detail. They then employed the alpha–omega formulation exclusively. As discussed below, however, such a comparison is itself conditioned by the continuity equation used by the model to determine . In the most recent multi-reanalysis assessment, Oh et al. (2026) likewise described the two formulations as theoretically equivalent when globally integrated, but noted that the alpha–omega formulation is more sensitive to uncertainty because of inaccuracies in reanalysis and therefore adopted the formulation.
More serious inconsistencies appear in the alpha–omega estimates themselves. From the conversion components reported by W.-M. Kim and Choi [10], the net APE-to-kinetic-energy conversion is for NCEP2 but for ERA-Interim. Independently, Makarieva et al. [18] likewise obtained negative global pressure-velocity power from NCAR/NCEP and positive from MERRA. Thus alpha–omega estimates can differ even in sign among reanalyses, whereas the horizontal-wind formulation is substantially more robust.
These results expose two distinct questions. First, are the alpha–omega and horizontal-wind formulations physically equivalent in a moist atmosphere? Second, what relation should be expected between the two diagnostics when they are evaluated from reanalysis fields whose mass-continuity equation, archived fields, and subsequent mass corrections need not reproduce physical gaseous-air continuity?
The first question is theoretical. The usual transformation between Eqs. (2) and (4) uses source-free continuity. In gaseous moist air, evaporation and condensation are sources and sinks of gas mass, and the vertical pressure gradient need not equal the weight of gaseous air alone. For a statistically stationary global control volume, we derive
where is the gravitational power associated with lifting water that subsequently leaves the gas phase and is the power associated with departure from gas-phase hydrostatic balance. The dry hydrostatic equality is recovered when both terms vanish.
The second question concerns the continuity equation actually associated with the reanalysis fields. Model pressure velocity is tied to the mass variable and source convention of the underlying dynamical system, which need not coincide with physical gaseous-air continuity. Archived fields can acquire additional continuity inconsistencies through assimilation, interpolation, and temporal sampling, while postprocessing procedures may impose physical dry-air conservation and reconstruct from the corrected mass fluxes. Agreement or disagreement between the two reanalysis diagnostics therefore need not constitute a direct measurement of the physical terms in Eq. (6).
This distinction is especially important because the global alpha–omega integral is a small residual of much larger positive and negative contributions. Steinheimer et al. [30] showed strong cancellation between these contributions, so spatially structured changes in can be small relative to the local terms yet large relative to the global residual.
Finite resolution introduces a further distinction between archived and total physical power. Pressure-level fields omit kinetic-energy generation by motions below their spatial and temporal resolution. They also omit the real partial layer between the lowest valid pressure surface and the ground. This can be consequential because prominent kinetic-energy generation occurs in the lower atmosphere [7], while our calculations below show a pronounced maximum near the surface.
Section 2 derives the relation between and for gaseous moist air, including nonhydrostatic vertical-force work. Section 3 examines how the mass-continuity conventions of reanalysis systems, continuity errors in archived fields, and subsequent mass corrections affect the relation between the two resolved diagnostics. Sections 4 and 5 evaluate both diagnostics from six-hourly ERA5, JRA-3Q, and MERRA-2 pressure-level fields for 2025. The analysis is designed to determine which differences between published power estimates reflect atmospheric physics and which reflect the construction and conditioning of the diagnostic.
Atmospheric power with phase changes and nonhydrostatic motion
The relation between Eqs. (2) and (4) follows from mass continuity and vertical force balance. Let be the gas-phase velocity, where is vertical velocity, is the upward unit vector, and is height. Since ,
where is total pressure-gradient power and is pressure storage. Splitting into horizontal and vertical parts gives
Gaseous moist air obeys
where is the gas-phase mass source, positive for evaporation and negative for condensation. Multiplication by and integration over the atmosphere gives
with
where contains gaseous gravitational-energy storage and boundary fluxes, is the control-volume boundary, and is its outward unit normal. For the global atmospheric volume, the boundary-flux term vanishes at the surface because and at the top of the atmosphere as atmospheric mass tends to zero.
For a statistically stationary global water cycle, let be the global-mean precipitation mass flux and define the effective source–sink height
Then
When evaporation occurs predominantly near the surface, can be interpreted as an effective mean height of removal of water from the gas phase.
Define the departure from gas-phase hydrostatic balance by
and let . Combining Eqs. (7), (8), and (10) gives
For a statistically stationary global control volume, , giving Eq. (6).
Here measures the global work done by departures from gas-phase hydrostatic balance. Condensate loading may contribute to this departure, but no decomposition of is required here. If while and is negligible, then ; this global condition does not imply negative buoyancy everywhere.
Pressure can be used as a vertical coordinate without assuming hydrostatic balance, provided that is monotonic with height. Define the pressure-to-height Jacobian
Then , and the volume-integral expansion power becomes
From Eq. (14),
Only under gas-phase hydrostatic balance, , does , reducing Eq. (17) to the conventional alpha–omega expression.
By contrast, the pressure-coordinate representation of does not require hydrostatic balance. Implicit differentiation of at fixed pressure gives
where denotes the horizontal gradient at fixed height. Since , the Jacobian cancels:
Therefore
The same distinction appears in mass continuity. In general pressure coordinates, define the mass Jacobian
(9) then becomes
Under hydrostatic balance, and , so that Eq. (23) reduces to
Vertical integration with consistent boundary terms then reproduces , the hydrostatic limit of Eq. (15).
Thus departures from hydrostatic balance affect the two pressure-coordinate diagnostics asymmetrically. They invalidate the conventional representation of by changing the pressure-to-volume Jacobian, whereas the corresponding Jacobian cancels identically in the pressure-coordinate representation of . The three reanalyses analyzed below employ hydrostatic dynamical cores, so this additional distinction does not affect the internal evaluation of their pressure-level diagnostics; it becomes relevant when these diagnostics are interpreted as measures of power in the actual atmosphere or applied to nonhydrostatic model output. Accordingly, these hydrostatic reanalyses cannot independently diagnose the actual atmospheric . In the comparisons below, enters through the physical identity for the atmosphere, not as a resolved reanalysis diagnostic.
Hydrostaticity therefore removes the Jacobian ambiguity for the reanalyses, but it does not remove a separate numerical issue: the alpha–omega diagnostic is strongly cancellation dominated. This follows from the same vertical pressure-work decomposition used above. Under hydrostatic balance,
Thus contains the mass-specific vertical pressure-gradient work , which changes sign between ascent and subsidence. Yet Eq. (10) constrains its global mass-weighted integral to . Large positive and negative vertical-work contributions must therefore nearly cancel in the global mean. This cancellation is distinct from the physical difference between and in Eq. (15); it persists even in a dry hydrostatic atmosphere.
The severity of this cancellation reflects the disparity between vertical and horizontal pressure gradients. A characteristic vertical gradient is
where is a pressure scale height, whereas a large-scale horizontal pressure gradient is typically of order . Even with , local vertical pressure-gradient power density can exceed its horizontal counterpart by about two orders of magnitude.
Unresolved motions and resolved mass continuity
The preceding section showed that, in the physical atmosphere, Eq. (15) relates to , , and the storage and boundary term . Since Eq. (11) expresses as the geopotential moment of the gas-phase mass source , the source or sink in gaseous mass continuity enters the difference between the two power formulations directly. In the hydrostatic limit, the physical difference between and therefore depends on this mass source, apart from storage and boundary terms.
The pressure velocity in a model or reanalysis, however, is tied to the continuity equation of its resolved dynamical system rather than directly to Eq. (24). To distinguish the two, let a superscript r denote resolved model or reanalysis fields, without implying a Reynolds averaging operation, and write their hydrostatic pressure-coordinate continuity equation schematically as
where is the source term represented in the continuity equation that determines the resolved mass transport. If the mass variable were gaseous air and its physical source were represented, then . More generally this need not be the case.
For a hydrostatic set of resolved fields whose pressure, mass measure, velocity, and continuity equation are internally consistent, the same integration used in § 2 gives
where denotes the corresponding storage and boundary contributions. Only when does have the meaning of a resolved gaseous . In particular, a source-free model continuity equation has and, apart from boundary and storage terms, restores the conventional hydrostatic equivalence of the two resolved diagnostics.
The distinction is explicit in the ECMWF Integrated Forecasting System (IFS). In the standard IFS formulation used before Cycle 46R1, the dynamical continuity equation conserves total atmospheric mass rather than dry-air mass [19]. Physical parameterizations can nevertheless change total atmospheric water through surface exchange, precipitation, and subgrid transport. Because these water-mass changes are not coupled to the standard dynamical continuity equation, they imply equal and opposite changes of dry-air mass, so that the total atmospheric mass source remains zero. Writing for such a total-water source or sink, the implied constituent bookkeeping is schematically
Internal phase changes among the model water species cancel in the total-water budget, so Eq. (29) should not be interpreted as an artificial dry-air source accompanying every condensation event. Its relevance here is that the source-free total-mass continuity equation used by the dynamical core does not contain the physical gas-phase source in Eq. (24).
This distinction provides a natural interpretation of the equivalence reported by Steinheimer et al. [30]. Their calculations used IFS Cycle 32R3; for selected cases they evaluated both the and formulations and found their global values identical within error margins. Since the dry-mass-conserving IFS continuity formulation was introduced only in Cycle 46R1 [19], their Cycle 32R3 calculation used the earlier source-free total-mass formulation. We infer that the reported agreement is consistent with the model’s source-free total-mass continuity constraint, provided that the pressure-work variables and mass measure used in the comparison were internally consistent with that continuity equation. It therefore does not establish physical gas-phase equivalence of the two formulations. ERA5 was produced with IFS Cycle 41R2 [6] and likewise predates the dry-mass-conserving formulation.
JRA-3Q has a closely related structure. It is based on a TL479 version of JMA’s global numerical weather prediction system as of December 2018 [13]. In the JMA Global Spectral Model, the specific-humidity and cloud-water equations contain tendencies from parameterized processes, whereas the pressure-mass continuity equation has a zero right-hand side. Pressure velocity is obtained by integrating this source-free continuity equation [8]. Thus model physics can change atmospheric water variables without the corresponding gas-phase source appearing in the continuity equation that determines . Accordingly, a model-output category such as “large-scale” or “resolved” precipitation is not equivalent to a resolved in Eq. (24).
MERRA-2 uses a different mass-budget convention. GEOS-5 was modified so that sources and sinks of atmospheric water enter the atmospheric mass budget while globally integrated dry atmospheric mass is conserved (Takacs, Suárez, & Todling, [31]). This avoids the compensating dry-air variations implied when total atmospheric mass is held fixed despite changes in water mass. Nevertheless, this is a continuity budget for total model atmospheric mass, not the gas-phase continuity equation of Eq. (9). In particular, phase changes between water vapor and suspended cloud condensate do not change total atmospheric mass, whereas conversion of suspended condensate to precipitation and re-evaporation of precipitation do enter the MERRA-2 mass source (Takacs et al., [31]). The three reanalyses therefore need not encode the physical gaseous source in the same way, even before any further processing of their archived fields.
The distinction between physical and model continuity is separate from another well-known problem: archived analysis fields need not satisfy the intended continuity constraint exactly. Such inconsistencies can arise from data assimilation, interpolation from model to pressure levels, temporal sampling, lower-boundary treatment, and other processing. Trenberth [32] found large continuity residuals in pressure-level ECMWF analyses and showed that postprocessing of the analyzed fields can substantially disrupt mass consistency. Subsequent work developed corrections of the analyzed velocity field designed to restore mass conservation ([34, 33]).
The target of these corrections is physically motivated rather than an arbitrary numerical closure condition. In the usual approximation that suspended condensate is neglected in the column mass budget, define the column masses of water vapor and dry air by
and the corresponding horizontal mass fluxes by
Physical dry-air conservation and the atmospheric water budget require
Adding the two equations gives the corresponding physical column mass budget,
Thus dry-air mass is conserved, whereas total atmospheric mass changes with the net surface exchange of water. The dry-air constraint can be imposed without prescribing and themselves; together with the moisture budget it makes the inferred mass balance consistent with, and testable against, independent estimates of .
This distinction is especially relevant when the native model and the postprocessed analysis obey different constraints. A field can be internally consistent with a model continuity equation that conserves total mass and yet fail the physical dry-air constraint in Eq. (32). Conversely, even a model with a physically consistent treatment of dry and water mass can acquire additional continuity residuals through assimilation and archival processing. A diagnosed continuity residual therefore need not have a single origin.
For uncorrected analysis fields, let
where the superscript a denotes the original analysis. A common correction is to add a height-independent horizontal velocity so that the corrected dry-air flux satisfies Eq. (32). For such a barotropic correction,
Equation (35) represents a simple column-barotropic correction; more complete three-dimensional adjustment procedures have also been developed [34]. The column continuity constraint fixes the divergence of the required mass-flux correction, but it does not uniquely specify a three-dimensional correction to the wind field. The choice of a barotropic correction, a vertically varying correction, or additional constraints on the correction therefore introduces information beyond mass conservation itself [32, 34].
Pressure velocity can subsequently be reconstructed by vertically integrating the local continuity equation satisfied by the corrected fields. In general, if the imposed pressure-coordinate continuity has source ,
The column correction makes the upper and lower boundary conditions compatible, but the resulting depends on both the continuity equation being imposed and the procedure used to correct the horizontal mass fluxes.
The role of such a correction is explicit in the calculation of Laliberté et al. [14]. Their remapping of CESM fields and data assimilation in MERRA left small continuity errors that could become significant in energy-flux calculations. They therefore applied a barotropic correction to the horizontal winds and recovered by integrating the corrected continuity equation, imposing vanishing at both the upper and lower boundaries. Their diagnostic is consequently a quantity conditioned by the imposed mass-continuity procedure rather than a diagnostic based directly on an independently archived pressure-velocity field.
There are therefore two conceptually distinct ways in which reanalysis fields can depart from the physical gaseous continuity equation used in § 2. First, the underlying model can employ a mass variable and source convention that differ from physical gaseous-air continuity, as in the source-free total-mass formulations discussed above. Second, assimilation, interpolation, temporal sampling, lower-boundary treatment, and other processing can introduce additional continuity residuals into the archived fields. A subsequent mass correction may restore an intended physical conservation constraint, but it does not uniquely reconstruct the unknown three-dimensional error in the original velocity field.
These distinctions are especially important for . As shown in § 2, the alpha–omega diagnostic is strongly cancellation dominated, so comparatively small structured changes in can produce substantial changes in its global integral. Corrections to horizontal winds can also alter , but is not reconstructed by vertically integrating a continuity residual and is not intrinsically subject to the same ascent–subsidence cancellation. Consequently, the reanalysis difference can reflect the mass-continuity convention of the underlying model, residual inconsistencies in archived fields, and any subsequent mass correction, in addition to the physical terms appearing in Eq. (15). Depending on how continuity and pressure velocity are treated, the resulting , and especially its difference from , can therefore vary substantially among reanalyses. This behavior is examined directly for ERA5, JRA-3Q, and MERRA-2 in the following section.
Data and methods
We use MERRA-2 [4], ERA5 [6], and JRA-3Q [13]. Their different models, assimilation systems, grids, and vertical discretizations provide a test of inter-reanalysis robustness. We analyze the full calendar year 2025 at 1460 common six-hourly times, applying identical processing to all three reanalyses for a structural intercomparison of the two power diagnostics.
We use pressure-level zonal and meridional wind, geopotential or geopotential height, pressure vertical velocity, temperature, and specific humidity, together with surface pressure. The files analyzed here contain 37 pressure levels from 1000 to 1 hPa for ERA5, 42 levels from 1000 to 0.1 hPa for MERRA-2, and 45 levels from 1000 to 0.01 hPa for JRA-3Q. Product-specific fill values are converted to missing values before a common physical mask is applied.
Because the three products extend to different upper pressure levels, we also tested a common upper boundary at 1 hPa. Truncation at 1 hPa changes by 0.008 W m for MERRA-2 and 0.031 W m for JRA-3Q, and changes by 0.002 and 0.010 W m, respectively. These differences do not affect the conclusions.
Moist specific volume is calculated from virtual temperature,
where is temperature, is specific humidity, is pressure in pascals, and is the gas constant of dry air. The local alpha–omega power per unit mass is
On each reanalysis latitude–longitude grid, kinetic-energy generation by horizontal pressure gradients per unit mass is
where and are zonal and meridional wind, is Earth’s radius, is longitude, is latitude, and is geopotential. Longitude derivatives use a periodic centered difference. For ERA5 and MERRA-2, centered latitude differences are used and the geographic pole rows are excluded. JRA-3Q is supplied on nonuniform Gaussian latitudes; its meridional derivative uses a second-order three-point formula on the actual latitude coordinates, with one-sided three-point formulas at the two polewardmost Gaussian rows. A value of is retained only when the central point and all four derivative-stencil points are finite and above the local surface. Geopotential height, where supplied, is converted using .
For , this five-point requirement can raise the effective lowest usable pressure level above that determined by the local surface pressure alone. A central point can itself lie above the surface while one or more neighboring points required for the derivative lie below terrain. This additional masking does not arise for , whose evaluation requires no horizontal derivative stencil. For all three reanalyses, we quantify the stencil effect by recording, at each pressure level, the area-weighted fraction for which the central point is valid, the fraction for which the full five-point stencil is valid, their difference, and the fraction of center-valid area retained by the stencil. The additional global area excluded by the stencil is confined mainly to the lowest pressure levels: its maximum annual-mean value is 2.1% in ERA5, 2.7% in JRA-3Q, and 3.6% in MERRA-2, in all cases at 975 hPa. At 1000 hPa, the stencil retains 97.0%, 95.9%, and 94.4% of the center-valid area, respectively. The additional loss falls below 0.5% of global area by 750 hPa and is negligible by 500 hPa. These quantities are averaged over the 1460 six-hourly analysis times.
For either or , complete archived layers are integrated as
where is the number of ordered pressure levels and is the level index. A layer is used only when both bounding values are finite and both pressures satisfy .
For , the pressure-level and stencil masks leave an unresolved interval between the lowest valid level and surface pressure . We estimate the sensitivity to this interval in two ways. First, as a zero-surface sensitivity construction, we set and connect this value linearly to the lowest valid value , giving
This construction provides a simple zero-surface treatment against which the linear extrapolation below can be compared.
As a second estimate, we linearly extrapolate from the two lowest valid levels, ,
and calculate
Where a second valid lower level is unavailable, the zero-surface estimate (41) is used instead. The two treatments characterize the sensitivity of to the unresolved near-surface layer without assuming that either reproduces its detailed vertical structure.
We do not include an analogous surface-layer correction in the reported . Unlike , the alpha–omega power density has no natural zero boundary value. At the lower boundary, pressure velocity satisfies
and is therefore generally nonzero at the surface. Consequently, has no corresponding natural zero-surface prescription. Nor is there a physical reason to assume that the strongly sign-varying field varies linearly from its lowest archived pressure level to the surface. We therefore report for complete archived pressure layers only, avoiding an additional lower-boundary assumption in this already cancellation-dominated diagnostic. Alternative surface constructions, where calculated, are retained only as numerical sensitivity tests and do not define the reported .
Annual global means are obtained by summing the vertically integrated column contributions with grid-cell-area weighting and dividing by the total surface area ; the normalization is therefore not changed by terrain or stencil masking. Masked contributions are treated as unresolved rather than by renormalizing the remaining valid area. For JRA-3Q, grid-cell areas are evaluated using the supplied Gaussian quadrature weights. Zonal means are formed analogously within each latitude band. Vertical profiles are annual global means of the archived layer contributions. Robustness is assessed from synchronized six-hourly correlations, the partition above and below 500 hPa, and signed latitude profiles.
OpenAI ChatGPT was used as an interactive assistant for manuscript editing, checking mathematical derivations and analysis code, and reference verification.
Results
Global integrals
Table 1 summarizes the 2025 annual means. On complete archived layers, , 2.48, and 2.84 W m in ERA5, JRA-3Q, and MERRA-2, respectively. With the zero-surface treatment , the omitted surface layer adds 0.23, 0.24, and 0.27 W m, giving totals of 2.59, 2.72, and 3.12 W m and a cross-reanalysis mean of 2.81 W m. Linear extrapolation gives larger surface contributions of 0.51, 0.54, and 0.63 W m, corresponding to totals of 2.87, 3.02, and 3.47 W m and a cross-reanalysis mean of 3.12 W m.
For comparison, Table 1 also gives the 1980–2023 mean resolved values calculated from the conversion components reported by Oh et al. (2026). They agree closely with the 2025 archived-layer values obtained here for all three reanalyses.
The alpha–omega diagnostic varies more strongly. Its archived-layer means are 3.23, 1.87, and 2.94 W m, while the corresponding values after linear surface extrapolation are 2.53, 1.28, and 3.20 W m. The cross-reanalysis range therefore increases from 1.36 to 1.92 W m. Pairwise correlations of the synchronized six-hourly series are 0.92–0.96 with and 0.93–0.96 with linear extrapolation, compared with 0.69–0.77 for linearly extrapolated .
Latitudinal structure and cancellation
The annual zonal-mean profiles reveal the contrasting numerical conditioning (Figure 1). Kinetic-energy generation by horizontal pressure gradients is positive through almost the entire latitude range and closely agrees among the three products. After area weighting, its positive latitude-band contributions are 2.37–2.85 W m, while negative contributions have magnitudes below 0.01 W m. Thus global is not a small residual of large opposite-signed zonal terms.
Figure 1. Annual zonal-mean archived-layer column power for 2025. (a) Kinetic-energy generation by horizontal pressure gradients, . (b) Alpha–omega power . Profiles are shown for ; the omitted polar bands make net signed contributions smaller than 0.01 W m to and 0.04 W m to . Note the different ordinate scales.
The alpha–omega profiles have similar large-scale shapes but much larger amplitudes. Column values reach about 390–410 W m near 7°N and −163 to −172 W m near 25–26°N, compared with global means of only 1.87–3.23 W m. Area-weighted positive contributions are 50.4–52.4 W m and negative contributions are −47.2 to −50.5 W m. The sum of the absolute positive and negative branch contributions therefore exceeds the global residual by factors of about 30–55.
Vertical structure and surface sensitivity
Horizontal kinetic-energy generation is concentrated in the lower atmosphere (Figure 2). The largest archived layer occurs at 962.5 hPa in all three products. Below 500 hPa, the archived atmosphere supplies 82.7%, 82.1%, and 72.5% of in ERA5, JRA-3Q, and MERRA-2. The corresponding values are 1.95, 2.04, and 2.06 W m, a range of only 0.11 W m.
Figure 2. Annual global-mean archived-layer contributions for 2025. (a) and (b) . Pressure decreases upward; the dashed line marks 500 hPa. Each point is the contribution of one pressure layer. Surface corrections are not included in the plotted profiles and are reported separately in Table 1.
The zero-surface treatment adds 0.23, 0.24, and 0.27 W m to in ERA5, JRA-3Q, and MERRA-2, respectively. These corrections are 9.6–9.7% of the archived-layer values and about 8.8% of the corrected totals. Linear extrapolation adds 0.51, 0.54, and 0.63 W m, raising the archived-layer values by 21.5–22.1% and accounting for 17.7–18.1% of the extrapolated totals. The difference between the two treatments is 0.28–0.35 W m across the three products, while their similar relative contributions indicate that the sensitivity is systematic rather than reanalysis specific.
Alpha–omega power is less coherent in the lower atmosphere. Its contribution below 500 hPa is 2.06 W m in ERA5, 0.76 W m in JRA-3Q, and 1.23 W m in MERRA-2. The contrast is even clearer in the surface treatment. Linear extrapolation consistently increases in all three reanalyses, by 0.51, 0.54, and 0.63 W m in ERA5, JRA-3Q, and MERRA-2, respectively. Applied in exactly the same way to , however, the surface correction is −0.70, −0.59, and +0.26 W m, respectively: it decreases in ERA5 and JRA-3Q but increases it in MERRA-2. Thus the near-surface correction has a consistent sign and magnitude for , but is sign-indefinite for .
On complete archived layers, is +0.87, −0.61, and +0.10 W m in ERA5, JRA-3Q, and MERRA-2, respectively. For the closure comparison, applying the common linear surface extrapolation to both diagnostics changes these differences to −0.34, −1.74, and −0.27 W m. The sign-indefinite alpha–omega surface contribution therefore adds substantial uncertainty to any closure calculation; the archived and extrapolated differences are interpreted separately below.
Discussion and conclusions
The archived-layer differences are 0.87 W m for ERA5, 0.10 W m for MERRA-2, and −0.61 W m for JRA-3Q. If these quantities were evaluated from a single internally consistent representation of physical gaseous-air continuity, Eq. (15) would relate their difference to the physical , , and boundary and storage terms. § 3 shows why this interpretation cannot in general be applied to reanalysis pressure-level fields.
The three reanalyses do not encode atmospheric mass continuity in the same way. ERA5 inherits the source-free total-mass continuity formulation of the pre-Cycle-46R1 IFS, while the JMA dynamical system underlying JRA-3Q likewise diagnoses pressure velocity from a source-free pressure-mass continuity equation. MERRA-2 instead uses a mass-budget formulation in which changes of atmospheric water affect total atmospheric mass while dry-air mass is conserved. Moreover, archived pressure-level fields can contain additional continuity inconsistencies introduced by assimilation, interpolation, temporal sampling, and lower-boundary treatment. The three values of above should therefore not be interpreted as three independent estimates of the physical .
This also changes the interpretation of agreement between the two diagnostics. The near equality of and in MERRA-2 does not imply that the physical is small. Conversely, the substantial positive ERA5 difference and negative JRA-3Q difference do not imply corresponding physical source or nonhydrostatic terms. Likewise, the agreement reported by Steinheimer et al. [30] for an earlier IFS cycle is naturally consistent with the source-free total-mass continuity equation used by that model generation and therefore does not establish physical equivalence of the two formulations.
Treatment of the unrepresented surface layer provides an additional numerical distinction. The alpha–omega surface extrapolation is not robust across the three reanalyses and makes negative in all three products. We therefore do not use the surface-extrapolated alpha–omega values to infer a physical closure. For , we instead carry two surface treatments. Imposing the zero-surface condition gives corrections of 0.23–0.27 W m and a cross-reanalysis mean total of 2.81 W m, while linear extrapolation gives 0.51–0.63 W m and a mean total of 3.12 W m. Thus the treatment of the omitted layer shifts the inferred mean by about 0.31 W m but does not alter its inter-reanalysis coherence or its order of magnitude. We retain both surface treatments when estimating total atmospheric power below.
Although the present analysis is restricted to 2025, its resolved values closely reproduce the 1980–2023 values calculated from the conversion components reported by Oh et al. (2026) (Table 1), with differences below 4% for all three reanalyses. This supports the use of the 2025 values as representative of the longer-term magnitude, while the present calculation itself does not quantify interannual variability.
| Reanalysis | archived | Oh et al. | sum | sum | archived | sum | |||
| MERRA-2 | 2.84 | 2.87 | 0.27 | 3.12 | 0.63 | 3.47 | 2.94 | 0.26 | 3.20 |
| ERA5 | 2.36 | 2.45 | 0.23 | 2.59 | 0.51 | 2.87 | 3.23 | −0.70 | 2.53 |
| JRA-3Q | 2.48 | 2.46 | 0.24 | 2.72 | 0.54 | 3.02 | 1.87 | −0.59 | 1.28 |
Table 1. Atmospheric-power diagnostics for 2025 from 1460 six-hourly analysis times. “Archived” denotes complete pressure layers. For , and sum use the zero-surface condition , while and sum use linear extrapolation to . For , only the linear surface treatment is shown. The column labelled “Oh et al.” gives the 1980–2023 mean resolved obtained as by Oh et al. (2026); these values do not include a surface correction. Values are W m.
The cancellation shown in Figure 1 helps explain why is especially sensitive to these distinctions. Its global value is a small residual of much larger positive and negative contributions, whereas is positive at almost every latitude and agrees much more closely among the three products. Trenberth-type mass corrections impose a physically motivated dry-air conservation constraint on analyzed fields, but the column constraint does not uniquely reconstruct the three-dimensional velocity error. When is subsequently recovered by vertically integrating the corrected continuity equation, the resulting field is conditional on both the continuity constraint and the chosen mass-flux correction. Corrections to horizontal winds can also alter , but is not itself reconstructed by vertical integration of continuity and is not intrinsically subject to the same ascent–subsidence cancellation.
Consequently, reanalysis-derived is best regarded as a diagnostic of the particular mass-continuity treatment embodied in, or subsequently imposed on, the analyzed fields rather than as an independent measurement of physical atmospheric power. Its difference from can contain the effects of model mass-continuity conventions, continuity residuals in archived fields, and postprocessing corrections in addition to the physical terms in Eq. (15). The large spread of across the three reanalyses is consistent with this sensitivity.
An observational scale for total can be obtained from the tropical precipitation-profile estimate. Pauluis and Dias [26], with the correction of Pauluis and Dias [27], estimated precipitation-induced dissipation at about 1.5 W m for the tropical belt 30°S–30°N from TRMM precipitation profiles for 1999–2007. Since this belt occupies one half of Earth’s surface, the tropical contribution alone corresponds to 0.75 W m in the global mean. Using GPCP v2.3 for 2000–2009, Nützel et al. [22] report mean precipitation rates of 3.05 mm day in the same tropical belt and 2.70 mm day globally, implying an extratropical mean of 2.35 mm day. If extratropical precipitation were assigned the same mean lifting height as tropical precipitation, the corresponding global would be about 1.33 W m. Neglecting the extratropical contribution altogether gives 0.75 W m. These deliberately crude estimates place global at order 1 W m. We therefore use W m only as an illustrative scale and emphasize the need for a dedicated global precipitation-path estimate.
To our knowledge, global is not presently constrained by observations. If , the two surface-corrected estimates of resolved , together with the illustrative physical W m, give cross-reanalysis total-power scales of about 3.8 W m for and 4.1 W m for linear extrapolation. Importantly, this inference uses and the independently estimated physical ; it does not use either reanalysis or the residual .
This power scale sharpens the contrast with estimates of generation within the Lorenz framework. Romanski and Rossow [29], using largely observation-based temperature and diabatic-heating fields, reported annual global means W m and W m. Their sum gives W m. To our knowledge, this remains the only global estimate in which APE generation was directly constrained by observations rather than inferred from reanalysis or model energetics. The revised radiative-flux-profile product of Zhang and Rossow [36] now makes such an updated calculation possible, but it has not yet been reported. The Romanski–Rossow estimate therefore remains a relevant observational benchmark.
A similarly low power scale was obtained by Kleidon [12], whose two-box maximum-power model gives about 1.6 W m for kinetic-energy generation associated with poleward heat transport between tropical and extratropical regions. This value should not, however, be interpreted as a general upper bound on kinetic-energy generation by differential heating. Kleidon’s model explicitly excludes kinetic-energy generation by the Hadley circulation, whereas in the conventional Lorenz energy cycle the Hadley circulation contributes to the conversion .
The more direct contrast for the present analysis is with . Even without a surface-layer correction, the resolved values of 2.36–2.84 W m substantially exceed the observationally constrained W m. This disparity suggests that a substantial part of kinetic-energy generation diagnosed by is not represented by conventional Lorenz APE generation. The two surface treatments raise the cross-reanalysis mean to 2.81–3.12 W m; including the illustrative W m gives the corresponding total atmospheric-power scale of about 3.8–4.1 W m.
This discrepancy motivates consideration of additional pathways of atmospheric power generation not represented by the conventional Lorenz APE framework. One such possibility has a long history. In 1885, Heinrich Hertz proposed an atmospheric steam-engine interpretation. As reported by Mulligan and Hertz [21], Hertz estimated the relevant evaporation–condensation temperature difference to be about 15 K. For K, this gives an efficiency scale . Applied to the global latent heat flux of order 80 W m corresponding to the GPCP precipitation rates quoted above, this gives a power scale of about 4 W m. The agreement in magnitude with the atmospheric power inferred here is consistent with a substantial role for water phase transitions in atmospheric power generation. A more complete thermodynamic formulation and quantitative assessment of this steam-engine argument are the subject of ongoing work. Resolving this discrepancy will require improved observational constraints, particularly on and , together with further theoretical work on atmospheric power-generation mechanisms.
Conflict of Interest
The authors declare there are no conflicts of interest for this manuscript.
Acknowledgments
This work received institutional support from the Biotic Pump Greening Group Institute and the Institute for Advanced Study, Technical University of Munich. No specific grant funding was received.
Data and Code Availability
ERA5 data were obtained from the Copernicus Climate Data Store, using the ERA5 hourly pressure-level and single-level products ([2, 3]). MERRA-2 data were obtained from the NASA Goddard Earth Sciences Data and Information Services Center, collection M2I3NPASM (inst3_3d_asm_Np) (Global Modeling and Assimilation Office (GMAO), 2015). JRA-3Q data were obtained from the NSF NCAR Geoscience Data Exchange, dataset d640000, using the high-resolution isobaric analysis (anl_p) and surface analysis (anl_surf) products [9]. The derived data and Mathematica scripts used in this study are archived in Zenodo at https://doi.org/10.5281/zenodo.21966314 [35].
References
- [1]Boer, G. J., & Lambert, S. (2008). The energy cycle in atmospheric models. Climate Dynamics, 30, 371–390. doi: 10.1007/s00382-007-0303-4DOI
- [2]Copernicus Climate Change Service. (2023a). ERA5 hourly data on pressure levels from 1940 to present [Dataset]. Copernicus Climate Change Service (C3S) Climate Data Store (CDS). Retrieved from https://doi.org/10.24381/cds.bd0915c6 doi: 10.24381/cds.bd0915c6
- [3]Copernicus Climate Change Service. (2023b). ERA5 hourly data on single levels from 1940 to present [Dataset]. Copernicus Climate Change Service (C3S) Climate Data Store (CDS). Retrieved from https://doi.org/10.24381/cds.adbb2d47 doi: 10.24381/cds.adbb2d47
- [4]Gelaro, R., McCarty, W., Suárez, M. J., Todling, R., Molod, A., Takacs, L., … Zhao, B. (2017). The modern-era retrospective analysis for research and applications, version 2 (MERRA-2). Journal of Climate, 30, 5419–5454. doi: 10.1175/JCLI-D-16-0758.1DOI
- [6]Hersbach, H., Bell, B., Berrisford, P., Hirahara, S., Horányi, A., Muñoz Sabater, J., … Thépaut, J.-N. (2020). The ERA5 global reanalysis. Quarterly Journal of the Royal Meteorological Society, 146, 1999–2049. doi: 10.1002/qj.3803DOI
- [7]Huang, J., & McElroy, M. B. (2015). A 32-year perspective on the origin of wind energy in a warming climate. Renewable Energy, 77, 482–492. doi: 10.1016/j.renene.2014.12.045DOI
- [8]Japan Meteorological Agency. (2019). Outline of the operational numerical weather prediction at the japan meteorological agency (Tech. Rep.). Tokyo, Japan: Japan Meteorological Agency. (188 pp.)
- [9]Japan Meteorological Agency. (2023). Japanese reanalysis for three quarters of a century (jra-3q) [Dataset]. Research Data Archive at the National Center for Atmospheric Research, Computational and Information Systems Laboratory. Retrieved from https://doi.org/10.5065/AVTZ-1H78 doi: 10.5065/AVTZ-1H78
- [10]Kim, W.-M., & Choi, Y.-S. (2017). Long-term change of the atmospheric energy cycles and weather disturbances. Climate Dynamics, 49(9–10), 3605–3617. doi: 10.1007/s00382-017-3533-0DOI
- [11]Kim, Y.-H., & Kim, M.-K. (2013). Examination of the global Lorenz energy cycle using MERRA and NCEP-reanalysis 2. Climate Dynamics, 40, 1499–1513. doi: 10.1007/s00382-012-1358-4DOI
- [12]Kleidon, A. (2021). Physical limits of wind energy within the atmosphere and its use as renewable energy: From the theoretical basis to practical implications. Meteorologische Zeitschrift, 30(3), 203–225. doi: 10.1127/metz/2021/1062DOI
- [13]Kosaka, Y., Kobayashi, S., Harada, Y., Kobayashi, C., Naoe, H., Yoshimoto, K., … Onogi, K. (2024). The JRA-3q reanalysis. Journal of the Meteorological Society of Japan, 102(1), 49–109. doi: 10.2151/jmsj.2024-004DOI
- [14]Laliberté, F., Zika, J. D., Mudryk, L., Kushner, P. J., Kjellsson, J., & Döös, K. (2015). Constrained work output of the moist atmospheric heat engine in a warming climate. Science, 347(6221), 540–543. doi: 10.1126/science.1257103DOI
- [16]Lorenz, E. N. (1955). Available potential energy and the maintenance of the general circulation. Tellus, 7, 157–167. doi: 10.3402/tellusa.v7i2.8796DOI
- [18]Makarieva, A. M., Gorshkov, V. G., Nefiodov, A. V., Sheil, D., Nobre, A. D., & Li, B.-L. (2017). Quantifying the global atmospheric power budget. Atmospheric Chemistry and Physics Discussions. (Preprint; revision not accepted for publication in ACP) doi: 10.5194/acp-2017-17DOI
- [19]Malardel, S., Diamantakis, M., Agusti-Panareda, A., & Flemming, J. (2019). Dry mass versus total mass conservation in the IFS (ECMWF Technical Memorandum No. 849). Reading, UK: European Centre for Medium-Range Weather Forecasts. doi: 10.21957/s3ho2aq29
- [20]Marques, C. A. F., Rocha, A., Corte-Real, J., Castanheira, J. M., Ferreira, J., & Melo-Gonçalves, P. (2009). Global atmospheric energetics from NCEP reanalysis 2 and ECMWF–ERA40 reanalysis. International Journal of Climatology, 29, 159–174. doi: 10.1002/joc.1704DOI
- [21]Mulligan, J. F., & Hertz, H. G. (1997). An unpublished lecture by heinrich hertz: “on the energy balance of the earth”. American Journal of Physics, 65(1), 36–45. doi: 10.1119/1.18565DOI
- [22]Nützel, M., Stecher, L., Jöckel, P., Winterstein, F., Dameris, M., Ponater, M., . . . Kunze, M. (2024). Updating the radiation infrastructure in MESSy (based on MESSy version 2.55). Geoscientific Model Development, 17, 5821–5849. doi: 10.5194/gmd-17-5821-2024DOI
- [24]Oort, A. H. (1964). On estimates of the atmospheric energy cycle. Monthly Weather Review, 92(11), 483–493. doi: 10.1175/1520-0493(1964)092<0483:OEOTAE>2.3.CO;2doi.org/10.1175/1520-0493(1964)092<0483:oeotae>2.3.co;2
- [25]Pan, Y., Li, L., Jiang, X., Li, G., Zhang, W., Wang, X., & Ingersoll, A. P. (2017). Earth’s changing global atmospheric energy cycle in response to climate change. Nature Communications, 8, 14367. doi: 10.1038/ncomms14367DOI
- [26]Pauluis, O., & Dias, J. (2012). Satellite estimates of precipitation-induced dissipation in the atmosphere. Science, 335(6071), 953–956. doi: 10.1126/science.1215869DOI
- [27]Pauluis, O., & Dias, J. (2013). Corrections and clarifications. Science, 339(6117), 271–271. doi: 10.1126/science.339.6117.271-bDOI
- [28]Peixoto, J. P., & Oort, A. H. (1992). Physics of climate. New York: American Institute of Physics.
- [29]Romanski, J., & Rossow, W. B. (2013). Contributions of individual atmospheric diabatic heating processes to the generation of available potential energy. Journal of Climate, 26(12), 4244–4263. doi: 10.1175/JCLI-D-12-00457.1DOI
- [30]Steinheimer, M., Hantel, M., & Bechtold, P. (2008). Convection in Lorenz’s global energy cycle with the ECMWF model. Tellus A, 60(5), 1001–1022. doi: 10.1111/j.1600-0870.2008.00348.xDOI
- [31]Takacs, L. L., Suárez, M. J., & Todling, R. (2016). Maintaining atmospheric mass and water balance in reanalyses. Quarterly Journal of the Royal Meteorological Society, 142(697), 1565–1573. doi: 10.1002/qj.2763DOI
- [32]Trenberth, K. E. (1991). Climate diagnostics from global analyses: Conservation of mass in ECMWF analyses. Journal of Climate, 4(7), 707–722. doi: 10.1175/1520-0442(1991)004<0707:CDFGAC>2.0.CO;2doi.org/10.1175/1520-0442(1991)004<0707:cdfgac>2.0.co;2
- [33]Trenberth, K. E., & Fasullo, J. T. (2018). Applications of an updated atmospheric energetics formulation. Journal of Climate, 31(16), 6263–6279. doi: 10.1175/JCLI-D-17-0838.1DOI
- [34]Trenberth, K. E., Hurrell, J. W., & Solomon, A. (1995). Conservation of mass in three dimensions in global analyses. Journal of Climate, 8(4), 692–708. doi: 10.1175/1520-0442(1995)008(0692:COMITD)2.0.CO;2doi.org/10.1175/1520-0442(1995)008<0692:comitd>2.0.co;2
- [35]Makarieva, A., & Nefiodov, A. (2026). Data and code for “The Alpha-Omega Problem in the Assessments of Atmospheric Power” [Dataset]. Zenodo. Retrieved from https://doi.org/10.5281/zenodo.21966314 doi: 10.5281/zenodo.21966314
- [36]Zhang, Y., & Rossow, W. B. (2023). Global radiative flux profile data set: Revised and extended. Journal of Geophysical Research: Atmospheres, 128(5), e2022JD037340. doi: 10.1029/2022JD037340DOI

