Pith. sign in

REVIEW 2 major objections 5 minor 50 references

Batch 2: Definition of novel Weather & Climate Dwarfs

T0 review · 2 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Pith's one-line read This report defines four isolated weather and climate 'dwarfs' that carry a major share of forecast computational cost.

desk verdict A useful packaging of four known NWP kernels as HPC benchmarks, worth reviewing after the printed MPDATA equations are fixed. read the letter →

arxiv 1908.07040 v1 pith:W3OIPSG2 submitted 2019-08-16 cs.DC

classification cs.DC
keywords weatherandclimatedwarfsMPDATAmultigridpreconditionerGCRsolverACRANEB2radiationexascalecomputingperformanceportabilityfinitevolume
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This deliverable argues that four components of numerical weather and climate prediction models can be cleanly isolated as 'dwarfs': standalone, documented kernels that reproduce the science and the computational load of their parent models. The four are a multigrid-preconditioned elliptic solver, two versions of the MPDATA advection scheme (structured and unstructured mesh), and the ACRANEB2 radiation scheme. The claim is that these kernels are representative of the dominant cost patterns in forecast production, and that packaging them this way lets downstream work optimize each one for new hardware without modifying full weather models. The report therefore provides scientific descriptions, installation and testing procedures, and reference configurations with error norms to anchor reproducibility.

What carries the argument

The central objects are the four dwarf definitions themselves. Each is a reduced but faithful encoding: GCR(k) with operator preconditioning via horizontal Richardson iteration and vertical Thomas-algorithm tridiagonal solves, accelerated by a multigrid v-cycle on nested octahedral meshes; MPDATA's anti-diffusive pseudo-velocity correction in its infinite-gauge form that removes the transported field from denominators; the same MPDATA idea with flux-corrected-transport limiters on structured grids; and ACRANEB2's two-stream radiation with single-interval band structure and net-exchanged-rate decomposition, refactored so vertical levels are the innermost loop. What these do for the argument is turn full models into small, measurable, portable kernels whose time-to-solution and energy-to-solution can be compared across architectures.

What would settle it

Run the structured MPDATA gauge-sphere subdwarf's default solid-body rotation test on a 128x64xL grid for one revolution; if the implementation follows the printed Eq. (50) literally, the L2 error norm will not match the reference value in Smolarkiewicz and Rasch (1991) because the anti-diffusive term is not normalized by psi.

Watch

Extended reading notes

Core claim

The paper's central claim is that the four selected components are faithful dwarfs: standalone, installable, testable kernels that replicate both the science and the computational load of their parent model components. The multigrid dwarf solves a three-dimensional potential-flow problem on an octahedral sphere mesh using GCR(k) iteration with a multigrid v-cycle preconditioner that splits horizontal and vertical inversions. The two MPDATA dwarfs implement the infinite-gauge anti-diffusive pseudo-velocity correction for advection on unstructured and structured meshes. The ACRANEB2 dwarf packages a short-range radiation scheme with two-stream radiative transfer, refactored so that the vertical level loop is innermost. For each dwarf the report supplies scientific background, I/O layout, installation and testing procedures, and reference configurations whose error norms anchor reproducibility across compilers and hardware.

Load-bearing premise

The whole deliverable rests on the assumption that its printed equations faithfully specify the implemented algorithms, but the MPDATA equations in Sections 3.2 and 3.3 omit a division by the transported field, so the equations alone cannot be used to rebuild the code.

Editorial extensions

If this is right

  • The four dwarfs become available on the project's software platform, so optimization work can start refactoring and benchmarking without waiting for full model builds.
  • The multigrid dwarf allows testing whether multigrid-preconditioned GCR reduces iteration counts and wall time compared to the split time-step deflation preconditioner.
  • The two MPDATA dwarfs enable a direct comparison of optimization potential between structured and unstructured mesh implementations, with L2 and Linf error norms as reference anchors.
  • The ACRANEB2 dwarf provides a physics-package benchmark with heavy transcendental functions and complex loops, and its transt3 subroutine alone accounts for more than 80 percent of runtime, making it a focused target for accelerator porting.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the dwarf isolation is faithful, the same four kernels could be reused as standardized benchmarks by the wider weather and climate HPC community because they are self-contained and reference-testable.
  • The dwarf framing may understate time-to-solution because real model coupling, halo exchanges, and I/O are removed, so performance numbers from isolated dwarfs may not transfer directly to full-model performance and should be validated with end-to-end timings.
  • The missing normalization in the printed MPDATA pseudo-velocity suggests that documentation-only releases of numerical algorithms are risky: an implementation may pass tests, but the printed specification cannot be re-implemented independently.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. This ESCAPE project deliverable defines four weather/climate computational dwarfs: a multigrid-preconditioned GCR elliptic solver, an unstructured-mesh MPDATA advection scheme, a structured-mesh MPDATA advection scheme, and the ACRANEB2 radiation scheme. For each dwarf the paper gives a scientific description, equations, implementation notes, installation instructions, and testing guidance. The central claim, stated in the executive summary, is that these four components form key functional parts of prediction models both scientifically and in terms of computational cost, and that the dwarfs are ready for use in the downstream ESCAPE optimization workflow.

Significance. If the definitions are correct, the paper is a useful, citable description of stable benchmark components for weather and climate HPC work. Its strengths are the anchoring to external analytic test cases (solid-body rotation on the sphere, potential flow over a Gaussian hill), the absence of fitted parameters, and the clear separation between scientific definition and packaging/installation procedure. The paper is honest about being a definition deliverable rather than a validation study, and it names its external references. However, because the code is not part of the manuscript and at least one central equation is internally inconsistent, the 'ready for use' claim is not currently established by the text alone.

major comments (2)
  1. [§3.2.3.2, Eq. (34); §3.3.3.3, Eq. (50)] The infinite-gauge pseudo-velocity in Eq. (34) (and again in Eq. (50)) is obtained from the standard form in Eq. (32)/(45) by dropping the 1/psi factors. Yet the update equations (35)/(51) still write the anti-diffusive correction as -dt*d(v_ad*psi)/dx. With v_ad defined by Eq. (34), this correction is the bracket of Eq. (30) multiplied by an extra factor psi, not the bracket itself; for a field with physical units the terms are also dimensionally inconsistent. The definition can be repaired either by retaining the 1/psi factors in v_ad, or by defining v_ad as the anti-diffusive flux and changing Eqs. (35)/(51) to use -dt*d(v_ad)/dx. As printed, the two MPDATA dwarfs cannot be re-implemented from the text, which conflicts with the deliverable's claim that the dwarfs are ready for use. The pseudo-code in Section 3.3.3.4 suggests the implementation uses the standard form, other_than the printed equations, but the text should be corrected.
  2. [§3.1.5, §3.2.4, §3.3.4, §3.4.4] The manuscript repeatedly states that the dwarf software has been made available, but the code is not part of the submission and the provided repository address (https://git.ecmwf.int/projects/ESCAPE) is an access-restricted project server rather than a public archive. A reader of the arXiv version cannot verify that the dwarfs exist, that the documented tests pass, or that the implementations correspond to the corrected equations. For a definition deliverable whose central claim is readiness for downstream use, a stable public code release or DOI is needed.
minor comments (5)
  1. [§3.2 and §3.3] The equation numbering is duplicated between the two MPDATA sections (Eqs. (23)-(35) and Eqs. (36)-(51)), making cross-references such as 'Eq. (36)' ambiguous and requiring the reader to guess which section is meant.
  2. [§3.2.3.1] The phrase 'To solve (36) numerically' should refer to Eq. (23) in that section, and 'Using this in (41)' should refer to Eq. (28); the references are off by the duplicated numbering.
  3. [§3.3.3.3, Eq. (50)] In Eq. (50), the notation 'dvad' and the velocity symbol 'u' should be 'v_ad' and 'v' to match Eq. (34) and the rest of the section.
  4. [§3.4.3.4] The paragraph beginning 'loop indices IIDIA(IAUCR) and IFDIA(IAUCR) are also calculated' appears twice with only slight differences; the duplicate should be removed.
  5. [§3.2.1 and §3.3.1] The claim that MPDATA accounts for 25-34% of model runtime is made without a citation or measurement; please provide a reference or a qualifying statement that this is an unpublished internal measurement.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the four dwarfs are defined from established external algorithms and validated against independent analytic test cases, with no fitted parameter or self-citation serving as the load-bearing scientific content.

full rationale

This deliverable defines four computational dwarfs by packaging existing, published numerical methods. The multigrid solver is presented as an implementation of the GCR method following Eisenstat et al. and Smolarkiewicz/Margolin, with a test case (potential flow over a Gaussian hill) whose governing equations are stated independently. The two MPDATA dwarfs reproduce the standard derivation of the antidiffusive pseudo-velocity from the upwind scheme via Taylor expansion, then reference the multidimensional formulation to Kühnlein and Smolarkiewicz and to the published MPDATA literature. The ACRANEB2 radiation dwarf is a stand-alone packaging of a published radiation scheme, with validation and performance work referenced to Mašek et al., Geleyn et al., and a DMI technical report. None of these steps fits a parameter to data and then predicts that same data; no quantity called a prediction is forced by an earlier fit. The ESCAPE project self-references (D1.1, D1.5, D3.3, the software exchange platform) frame the deliverable within the project workflow, but they do not carry the scientific derivation of the dwarfs. The dimensional inconsistency noted in Eqs. (34) and (50), where the infinite-gauge pseudo-velocity omits the 1/psi division present in Eq. (32), is a correctness concern that would affect re-implementation, but it is not a circularity: it does not make the output equivalent to the input. The paper is therefore self-contained with respect to circularity, and no circular step can be quoted or exhibited. Score 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The report is a packaging of known algorithms, so the main unstated inputs are the representativeness of the chosen test cases, the correctness of the implementations, and the validity of prior published approximations; the document's own equations introduce one candidate error (Eq. 34), which makes the MPDATA definition unreliable on paper.

assumptions (4)
  • domain assumption The GCR(k) solver with the described deflation and multigrid preconditioner converges efficiently for the potential flow benchmark.
    Section 3.1.3 asserts the approach reduces iterations but provides no iteration counts or convergence plots; efficacy is assumed from Refs [2,3] and internal project work.
  • domain assumption The potential flow over a Gaussian hill in terrain-following coordinates is a suitable canonical test for NWP elliptic solvers.
    Section 3.1.1 and 3.1.3.4 adopt this test from Smolarkiewicz and Margolin without comparing against other benchmark problems.
  • ad hoc to paper The infinite-gauge MPDATA formulation can remove ψ from the denominator of the antidiffusive velocity without changing the form vad ψ.
    Eq. (34)/(50) is written without the denominator present in the standard derivation (Eq. 32); no justification or reference for this specific form is given.
  • domain assumption ACRANEB2's two-stream, single-interval spectral approximations are appropriate for short-range NWP radiation.
    Inherited from Masek et al. (2016) and Geleyn et al. (2017); the deliverable does not re-derive or validate these approximations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Batch 2: Definition of novel Weather & Climate Dwarfs." pith.science (2026). https://pith.science/paper/W3OIPSG2

@misc{pith2026190807040,
  author       = {Pith},
  title        = {Pith review of: Batch 2: Definition of novel Weather & Climate Dwarfs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W3OIPSG2}},
  note         = {Machine review of arXiv:1908.07040}
}
read the original abstract

This document is one of the deliverable reports created for the ESCAPE project. ESCAPE stands for Energy-efficient Scalable Algorithms for Weather Prediction at Exascale. The project develops world-class, extreme-scale computing capabilities for European operational numerical weather prediction and future climate models. This is done by identifying weather & climate dwarfs which are key patterns in terms of computation and communication (in the spirit of the Berkeley dwarfs). These dwarfs are then optimised for different hardware architectures (single and multi-node) and alternative algorithms are explored. Performance portability is addressed through the use of domain specific languages. This deliverable contains the description of the characteristics of a second set of so-called numerical weather & climate prediction dwarfs that form key functional components of prediction models in terms of the science that they encapsulate and in terms of computational cost they impose on the forecast production. The ESCAPE work flow between work packages centres on these dwarfs and hence their selection, their performance assessment, code adaptation and optimisation is crucial for the success of the project. These new dwarfs have been chosen with the purpose of extending the range of computational characteristic represented by the dwarfs previously selected in batch 1 (see Deliverable D1.1). The dwarfs have been made, their documentation has been compiled and the software has been made available on the software exchange platform. The dwarfs in this deliverable include a multigrid elliptic solver, a novel advection scheme for unstructured meshes, an advection scheme for structured meshes and a radiation scheme. This deliverable includes their scientific description and the guidance for installation, execution and testing.

Figures

Figures reproduced from arXiv: 1908.07040 by the authors.

Figure 1
Figure 1. The JLON loop cannot be seen here, as this is now moved to include a larger t f f bti Fig. 1. The JLON loop cannot be seen here, as this is now moved to include a larger JLON loop cannot be seen here, as this is now moved to include a larger part of of subroutine. [PITH_FULL_IMAGE:figures/full_fig_p046_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 50 canonical work pages

  1. [1]

    Eisenstat, H.C

    S.C. Eisenstat, H.C. Elman, and M.H. Schultz. Variational iterative methods for nonsymmetric systems of linear equations. SIAM Journal on Numerical Analysis, 20(2):345–357, 1983

  2. [2]

    Smolarkiewicz and L.G

    P.K. Smolarkiewicz and L.G. Margolin. Variational solver for elliptic problems in atmospheric flows. Appl. Math. Comp. Sci, 4(4):527–551, 1994

  3. [3]

    Smolarkiewicz and L.G

    P.K. Smolarkiewicz and L.G. Margolin. Variational methods for elliptic problems in fluid models. In Proc. ECMWF Workshop on Developments in numerical methods for very high resolution global models, pages 137–159, 2000

  4. [4]

    Smolarkiewicz and J.A

    P.K. Smolarkiewicz and J.A. Pudykiewicz. A class of semi-Lagrangian approximations for fluids. J. Atmos. Sci., 49:2082–2096, 1992

  5. [5]

    Elliptic problems in linear differential equations over a network

    L.H.Thomas. Elliptic problems in linear differential equations over a network. Watson Sci. Comput. Lab Report, Columbia University, New Youk, 1949. 17 ESCAPE - WP1: Weather & Climate Dwarfs - D1.2: Batch no. 2 of Dwarfs 3.2 MPDATA for unstructured meshes 3.2.1 Scope MPDATA stands for multidimensional positive definite advection transport algorithm. We have...

  6. [10]

    (13) The spatial derivatives are discretised with upwind fluxes which results in three dimensions in a sum of the upwind fluxes over all interfaces between neighbouring grid cells

    rhofac_correction : for all nodes: for all levels: compute product of density and advected field 3 MPDATA dwarf for unstructured meshes Descretising ( 9 ) with forward di erence in time gives  n +1 i =  n i ≠ ” t ˆ ˆ x ( v  ) ≠ ” t ˆ ˆ x ( v ad  ) . (13) The spatial derivatives are discretised with upwind fluxes which results in three dimensions in a su...

  7. [11]

    compute_upwind_flux : for all edges: for all levels: compute the upwind flux for that edge

  8. [12]

    compute_upwind_fluz : for all nodes: for all levels: compute the upwind flux between vertically neighbouring grid cells

Show all 50 references
  1. [13]

    compute_fluxzdiv : for all nodes: for all levels: compute sum over all horizontal and vertical fluxes weighted with the surface of the corresponding face

  2. [14]

    advance_solution : for all nodes: for all levels: add computed sum to the advected field pD multiplied with the timestep and devided through the density

  3. [15]

    In each step:

    rhofac_correction : for all nodes: for all levels: compute product of density and advected field 3 1+2 3 4 10+11 13 14 6+7 8 10+11 The numbers next to each bar stand for one of the functions of the dwarf as given by the following pseudo-code: main program: loop over all timeste...

  4. [16]

    compute_upwind_flux: for all edges: for all levels: compute the upwind flux for that edge

  5. [17]

    compute_upwind_fluz: for all nodes: for all levels: compute the upwind flux between vertically neighbouring grid cells

  6. [18]

    compute_fluxzdiv: for all nodes: for all levels: compute sum over all horizontal and vertical fluxes weighted with the surface of the corresponding face

  7. [19]

    2 of Dwarfs

    advance_solution: for all nodes: for all levels: add computed sum to the advected field pD multiplied with the timestep and divided through the density 20 ESCAPE - WP1: Weather & Climate Dwarfs - D1.2: Batch no. 2 of Dwarfs

  8. [20]

    rhofac_correction: for all nodes: for all levels: compute product of density and advected field

  9. [21]

    compute_centered_flux: for all edges: for all levels: compute centred (averaged) flux for that edge

  10. [22]

    compute_centered_fluz: for all nodes: for all levels: compute centred flux between vertically neighbouring grid cells

  11. [23]

    compute_fluxzdiv: for all nodes: for all levels: compute sum over all horizontal and vertical centred fluxes weighted with the surface of the corresponding face

  12. [24]

    halo_exchange: communicate results between neighbouring processors

  13. [25]

    compute_pseudovel_xy: for all edges: for all levels: compute first term in the pseudo-velocity for that edge for all edges: for all levels: compute second term in the pseudo-velocity and add to first term

  14. [26]

    compute_pseudovel_z: for all nodes: for all levels: compute first term in the pseudo-velocity between vertically neighbouring grid cells for all nodes: for all levels: compute second term in the pseudo-velocity and add to first term for vertically neighbouring grid cells

  15. [27]

    2 of Dwarfs

    compute_fluxzdiv: for all nodes: for all levels: compute sum over all horizontal and vertical antidiffusive fluxes weighted with the surface of the corresponding face 21 ESCAPE - WP1: Weather & Climate Dwarfs - D1.2: Batch no. 2 of Dwarfs

  16. [28]

    advance_solution: for all nodes: for all levels: add computed sum to the advected field pD multiplied with the timestep and divided through the density

  17. [29]

    SSH keys

    halo_exchange: communicate results between neighbouring processors 3.2.4 Dwarf usage and testing In this section we describe how to download and install the dwarf along with all its dependencies, and we show how to run it for a simple test case. Note that the MPDATA dwarf for ...

  18. [31]

    Monotone advection on the sphere: an Eulerian versus semi-Lagrangian approach

    P.K. Smolarkiewicz and P.J. Rasch. “Monotone advection on the sphere: an Eulerian versus semi-Lagrangian approach”. In:Journal of the Atmospheric Sciences(1990). 3.3 MPDATA for structured meshes 3.3.1 Scope MPDATA stands for multidimensional positive definite advection transpor...

  19. [32]

    libmpdata++ 1.0: a library of parallel MPDATA solvers for systems of generalised transport equations

    A. Jaruga et al. “libmpdata++ 1.0: a library of parallel MPDATA solvers for systems of generalised transport equations”. In:Geoscientific Model Development8.4 (2015), pp. 1005–1032.doi: 10 . 5194 / gmd - 8 - 1005 - 2015. url: https : / / www . geosci - model-dev.net/8/1005/2015/

  20. [33]

    An unstructured-mesh finite-volume MP- DATA for compressibleatmospheric dynamics

    C. Kühnlein and P.K. Smolarkiewicz. “An unstructured-mesh finite-volume MP- DATA for compressibleatmospheric dynamics”. In:Journal of Computational Physics (submitted 2016)

  21. [34]

    The multidimensional posi- tive definite advection transport algorithm: Nonoscillatory option

    Piotr K Smolarkiewicz and Wojciech W Grabowski. “The multidimensional posi- tive definite advection transport algorithm: Nonoscillatory option”. In:Journal of Computational Physics86.2 (1990), pp. 355–375

  22. [35]

    MPDATA: A Finite-Difference Solver for Geophysical Flows

    Piotr K. Smolarkiewicz and Len G. Margolin. “MPDATA: A Finite-Difference Solver for Geophysical Flows”. In:Journal of Computational Physics140.2 (1998), pp. 459–

  23. [37]

    Monotone advection on the sphere: An Eulerian versus semi-Lagrangian approach

    Piotr K Smolarkiewicz and Philip J Rasch. “Monotone advection on the sphere: An Eulerian versus semi-Lagrangian approach”. In:Journal of the Atmospheric Sciences 48.6 (1991), pp. 793–810. 36 ESCAPE - WP1: Weather & Climate Dwarfs - D1.2: Batch no. 2 of Dwarfs 3.4 ACRANEB2 radi...

  24. [38]

    What is the potential of refactoring the dwarf to run optimally on 2016 model Xeon, Xeon Phi processors and GPUs?

  25. [39]

    2 of Dwarfs

    Is it worthwhile refactoring a physics subroutine such as this? 37 ESCAPE - WP1: Weather & Climate Dwarfs - D1.2: Batch no. 2 of Dwarfs

  26. [40]

    Should the refactoring be done universally for the different hardware architectures investigated?

  27. [41]

    Can a set of rules be made for proper programming optimised for current hardware architectures? Detailed descriptions of the physics in ACRANEB2 have been made by Mašek et al. (2016)

  28. [42]

    NEB”), ice cloud load (“ICE

    and Geleyn et al. (2017) [3] for the shortwave and longwave radiation, respectively. The work package 2 (WP2) software adaptation of the ACRANEB2 dwarf, and the benchmarking and diagnostics results for different hardware architectures (WP3) are detailed in the report by Poulsen...

  29. [43]

    Establish a solid reference (test case and source code) that reproduces the necessary results

  30. [44]

    43 ESCAPE - WP1: Weather & Climate Dwarfs - D1.2: Batch no

    Establish build and run environment to ease repetition and reproducibility. 43 ESCAPE - WP1: Weather & Climate Dwarfs - D1.2: Batch no. 2 of Dwarfs

  31. [45]

    a SPMD approach

    Ensure proper threading, i.e. a SPMD approach. •This requires a transition to Fortran90 assumed shape and trimming stack memory usage, and •contiguous data

  32. [46]

    •Reducing the stack pressure by reducing local 2D/3D variables to 1D/2D vars or even scalars

    Strive towards a minimal implementation, including: •Reducing the memory overhead. •Reducing the stack pressure by reducing local 2D/3D variables to 1D/2D vars or even scalars. •The largest stack arrays should be moved to the heap with proper NUMA initialisation of these heap ...

  33. [47]

    Performance studies (ACRANEB2)

    Continued refactoring by shuffling computations around to maximize parallel expo- sure (playing with data structures and loops). •Identifying computational patterns, e.g. with reduction and prefix sums, and •...without (SIMD-suitable loops) dependencies. •Re-organising heavy loop...

  34. [48]

    The potential of refactoring ACRANEB2 to run optimally contemporary Xeon, Xeon Phi processors and GPUs is very large

  35. [49]

    We find it very likely that this is the case for legacy physics routines used in weather and climate models in general

  36. [50]

    For a complex physics subroutine, such as this, we found that different refactoring was needed for the GPUs and the Xeon Phi processors

  37. [51]

    3.4.7 References for ACRANEB2

    Rules can be made for how to adjust legacy physics code for optimised running on current hardware architectures, however, the work of specialists is required! Such an investment is worthwhile given the significant improvement in time-to-solution and reduction in energy consumpt...

  38. [52]

    Single interval shortwave radiation scheme with parameterized optical saturation and spectral overlaps

    J. Mašek et al. “Single interval shortwave radiation scheme with parameterized optical saturation and spectral overlaps”. In:Q. J. R. Met. Soc.142.694 (2016), pp. 304–326. doi: 10.1002/qj.2653

  39. [53]

    J. W. Poulsen and P. Berg.Tuning the implementation of the radiation scheme ACRANEB2. Tech. rep. 17-22. Danish Meteorological Institute, Copenhagen, Den- mark, 2017

  40. [54]

    Single interval longwave radiation scheme based on the net exchanged rate decomposition with bracketing

    J.-F. Geleyn et al. “Single interval longwave radiation scheme based on the net exchanged rate decomposition with bracketing”. In:Q. J. R. Met. Soc.143.704 (2017), pp. 1313–1335.doi: 10.1002/qj.3006. 47 ESCAPE - WP1: Weather & Climate Dwarfs - D1.2: Batch no. 2 of Dwarfs

  41. [55]

    A nonisothermal emissivity and absorptivity formulation for water vapor

    V. Ramanathan and P. Downey. “A nonisothermal emissivity and absorptivity formulation for water vapor”. In:J. Geophys. Res.91.D8 (1986), pp. 8649–8666.doi: 10.1029/JD091iD08p08649. 4 Conclusions A second set of weather and climate prediction model sub-components (dwarfs) has b...

  42. [480]

    doi: https : / / doi

    issn: 0021-9991. doi: https : / / doi . org / 10 . 1006 / jcph . 1998 . 5901. url: http://www.sciencedirect.com/science/article/pii/S0021999198959010

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.