Comparing the dynamics of idealized squall lines between NWP and LES models

arXiv:2506.19435 · physics.ao-ph, physics.flu-dyn · Submitted 2025-06-24 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: We're looking at the paper: "Comparing the dynamics of idealized squall lines between NWP and LES models".

Mira: Here's the gist of the paper. Both Numerical Weather Prediction (NWP) models and Large-Eddy Simulation (LES) models are used to simulate convective systems, such as squall lines, but with different purposes NWP models aim for the most accurate weather forecasts, whereas LES models are typically used to advance our understanding of physical processes.

Kai: With increasing computer power, the domain sizes and resolutions of these models converge, which raises the question if the model results also converge.

Mira: The study investigated an idealized squall line with the NWP model ICON (ICOsahedral Non-hydrostatic) and the LES model MicroHH.

Kai: However, there are quantitative differences with MicroHH having a more intense squall-line circulation than ICON at all resolutions (1 km, 500 m, and 250 m), mainly because MicroHH has less numerical diffusion.

Paper summary: Kai: So, we're looking at this paper today, "Comparing the dynamics of idealized squall lines between NWP and LES models." This study compares two different ways to simulate a squall line: a Numerical Weather Prediction model and a Large-Eddy Simulation model. The main idea here is to see if these two different tools, one for forecasting weather and one for understanding physics, agree when simulating the same thing.

Mira: Right, so the core thesis is that even though NWP models are for making forecasts and LES models are for understanding the underlying physics, they should still be able to show similar things about a convective system like a squall line if they're both set up correctly. The authors investigate what happens when you use ICON for NWP and MicroHH for LES.

Lev: From an engineering standpoint, it’s interesting because these models have fundamentally different designs. ICON uses compressible equations on a triangular grid, while MicroHH uses anelastic equations on a regular grid, which means they handle the physics in really different ways.

Kai: Exactly. So the paper sets up this idealized squall line comparison between ICON and MicroHH. They align the setup for things like resolution, domain size of two thousand five hundred ninety-two kilometers across the line direction, boundary conditions, and even microphysics schemes like warm and ice processes to make sure they are comparing apples to apples.

Mira: That's a key part of the methodology they used. They aimed for a good alignment between these different setups in both models so they could isolate what causes the differences in their outputs later on.

Lev: And when you look at the results, the paper points out that MicroHH has a more intense squall-line circulation than ICON across all resolutions—at one kilometer, five hundred meters, and two hundred fifty meters—and this difference is mainly because MicroHH has less numerical diffusion.

Kai: Less numerical diffusion is huge. That suggests that the way the model handles small-scale turbulence in MicroHH allows it to capture more of that intense circulation compared to ICON. That’s a direct link between the physics and the numerics they are using there, which is something we always worry about with computer power convergence.

Mira: It really highlights how much model physics matters when you're looking at these quantitative differences across different resolutions. Even though they see a good qualitative agreement in the structure and circulation patterns, those quantitative gaps show where the models aren't quite matching up perfectly yet.

Lev: That quantitative gap is where we need to focus on improving the advection scheme or the diffusion formulation, because sensitivity tests showed that different formulations of turbulent diffusion cause a two to eight percent change in mass flux.

Kai: So the authors found that varying the advection scheme can actually alter the mass flux by sixteen percent. That tells us that how they move things around—the advection—is really controlling how strong the updraft looks in these simulations.

Mira: It seems like when you look at warm microphysics, both models agree qualitatively on the squall-line structure, but they differ quantitatively in things like the strength of the updraft and the horizontal extent of the cloud.

Lev: In fact, for those runs with only condensation—meaning no rain processes—the mean vertical velocity and mass flux are higher in MicroHH compared to ICON. But they also noted that these differences between models get larger when you include condensation only than when you look at runs with warm microphysics.

Kai: That suggests that the rain processes, or the cooling from precipitation, play a role in dampening those differences we saw earlier, which points toward evaporation cooling being important for dynamics.

Mira: When they included ice microphysics, the qualitative agreement between MicroHH and ICON still held up regarding the squall-line structure itself. But then things got different in terms of how hydrometeors spread out.

Lev: Specifically, MicroHH resulted in a wider trailing stratiform region of moderate radar reflectivity, while ICON produced a higher maximum radar reflectivity and more precipitation right at the surface. That's a very concrete difference in the output you get from each model.

Kai: So to sum up this paper on "Comparing the dynamics of idealized squall lines between NWP and LES models," it shows that both NWP and LES can simulate the same system structure, but there are measurable quantitative differences, especially in rainfall amounts. The authors found these differences stem from things like advection, diffusion, and resolution settings.

Mira: It really emphasizes the importance of investing in developing the physics alongside improving the numerics because they demonstrated that good qualitative agreement between these two modeling approaches is possible under certain conditions.

Lev: What this means for us in terms of real hardware is that we need to be very careful about how we implement turbulent diffusion and advection schemes when we push towards higher resolutions, because those numerical choices have a measurable impact on the updraft dynamics.

Kai: So, looking ahead at what this study suggests, it points toward focusing our efforts on refining the physics and making sure the numerics are robust enough to handle those physical differences across different simulation setups.

Mira: It suggests that even as computer power increases and domain sizes get larger, we still have these specific physical and numerical hurdles to overcome when trying to perfectly replicate complex convective systems with either NWP or LES.

Conclusion: Kai: So, we've been looking at how ICON, an NWP model, compares to MicroHH, which is a Large-Eddy Simulation model, when they both try to simulate this squall line.

Mira: And the authors of this paper are Redelsperger and their team. They focused on comparing the dynamics of these two different modeling approaches across different resolutions like one kilometer, five hundred meters, and two hundred fifty meters.

Lev: The main thing they did was align the setup—the grid size, the boundary conditions, and even how they handled things like ice microphysics—so you could really isolate what caused the differences in their outputs.

Kai: And what they found is that while both models agree on the general shape of the squall line structure, there are still quantitative differences.

Mira: That's right. They saw things like a difference in updraft strength and how far out the cloud spreads when comparing warm microphysics runs between ICON and MicroHH.

Lev: The paper points out that these differences come down to things like turbulent diffusion and the advection scheme, with the numerical diffusion in those schemes having a real impact on how the updraft looks.

Kai: So what does this title 'Comparing the dynamics of idealized squall lines between NWP and LES models' actually mean for us?

Mira: It means that even when you’re running two fundamentally different types of models, one for forecasting weather and one for studying physics, if you get the setup right, they can actually show similar things about how these intense convective systems behave.

Lev: But those differences aren't just academic. The paper suggests that investing in both better physics schemes and better numerical methods is crucial because those are what cause those quantitative gaps we saw.

Kai: So, if the models agree qualitatively, but differ quantitatively on things like rainfall amount or circulation intensity across different resolutions, what does that tell us about the limitations of each approach?

Mira: It tells us that the assumptions underpinning each model—whether it's the anelastic equations in MicroHH or the compressible equations in ICON—are causing those discrepancies when you zoom in on smaller scales.

Lev: And from where I’m sitting, it suggests that for real hardware, we can’t just rely on one approach; you have to account for these numerical choices because they control the dynamics of things like mass flux.

Kai: So, the conclusion is that the good qualitative agreement between NWP and LES justifies assuming that insights from LES can be applied in NWP models, provided you understand where those quantitative differences are coming from.

Mira: Exactly. It's not a perfect match yet, but it shows that as computer power increases, we need to keep refining both the physics and the numerics to get closer to that ideal agreement.

Lev: And this leads us right into how those numerical choices, like advection schemes and diffusion formulations, can actually change results by up to sixteen percent in mass flux.

Mirjam Tijhuis, Axel Seifert, Alberto de Lozar, Bart J. H. van Stratum, Chiel C. van Heerwaarden

Wageningen University & Research · Deutscher Wetterdienst

physics.ao-ph, physics.flu-dyn

Submitted: 2025-06-24

Updated: 2026-11-04

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 76/100

The gist: Both Numerical Weather Prediction (NWP) models and Large-Eddy Simulation (LES) models are used to simulate convective systems, such as squall lines…

Key concepts

NWP Models (ICON)
Numerical Weather Prediction models aim for the most accurate weather forecasts. ICON uses compressible equations on a triangular grid, which is good for large scales but requires small time steps due to acoustic waves.
LES Models (MicroHH)
Large-Eddy Simulation models are used to understand physical processes. MicroHH solves anelastic equations on a regular grid and uses approximations to filter out acoustic waves, allowing for larger time steps compared to NWP models.
Numerical Diffusion
This refers to how numerical methods smooth out small-scale features in a simulation. The study found that the advection scheme's numerical diffusion significantly controls updraft dynamics, causing measurable changes in mass flux between the models.
Microphysics Schemes
Microphysics describes how water changes phase (e.g., liquid to ice). Differences in these schemes, such as warm versus ice microphysics, caused quantitative variations in cloud structure and precipitation patterns between the two models.

Terminology

Summary

Both Numerical Weather Prediction (NWP) models and Large-Eddy Simulation (LES) models are used to simulate convective systems, such as squall lines, but with different purposes NWP models aim for the most accurate weather forecasts, whereas LES models are typically used to advance our understanding of physical processes. p. 1

Where the paper says it

ABSTRACT: Both Numerical Weather Prediction (NWP) models and Large-Eddy Simulation (LES) models are used to simulate convective systems, such as squall lines, but with different purposes. NWP models aim for the most accurate weather forecasts, whereas LES models are typically used to advance our understanding of physical processes.

With increasing computer power, the domain sizes and resolutions of these models converge, which raises the question if the model results also converge. p. 1
Where the paper says it

Therefore, these types of models differ in their design. With increasing computer power, the domain sizes and resolutions of these models converge, which raises the question if the model results also converge.

The study investigated an idealized squall line with the NWP model ICON (ICOsahedral Non-hydrostatic) and the LES model MicroHH. p. 1
Where the paper says it

With increasing computer power, the domain sizes and resolutions of these models converge, which raises the question if the model results also converge. We investigated an idealized squall line with the NWP model ICON (ICOsahedral Non-hydrostatic) and the LES model MicroHH.

However, there are quantitative differences with MicroHH having a more intense squall-line circulation than ICON at all resolutions (1 km, 500 m, and 250 m), mainly because MicroHH has less numerical diffusion. p. 1
Where the paper says it

The models simulate the same squall-line structure and circulation pattern in simulations with both warm and ice microphysics. However, there are quantitative differences with MicroHH having a more intense squall-line circulation than ICON at all resolutions (1 km, 500 m, and 250 m), mainly because MicroHH has less numerical diffusion.

The quantitative differences between the models across resolutions highlight the importance of model physics and numerics, whereas the good qualitative agreement gives confidence that insights from LES can be applied in NWP. p. 1
Where the paper says it

The magnitude of the differences is sensitive to the advection scheme and the resolution and less sensitive to the formulation of turbulent diffusion. The quantitative differences between the models across resolutions highlight the importance of model physics and numerics, whereas the good qualitative agreement gives confidence that insights from LES can be applied in NWP.

Model Design Differences

The models differ in their design, mainly because ICON solves the compressible equations on a triangular grid, while MicroHH solves the anelastic equations on a regular grid. ICON uses compressible equations which are applicable over a large range of scales and potentially use unstructured grids to be applicable in spherical geometry Compressible equations are hyperbolic, requiring only local communication for scalability but include acoustic waves that require a short time step. p. 3

Where the paper says it

range of scales, and potentially the use of unstructured grids to be applicable in spherical geometry. Compressible equations are hyperbolic; hence, the propagation speed of disturbances is finite, and only local communication is required, which improves scalability.

In contrast, LES models are commonly designed on structured grids and use either the anelastic or Boussinesq approximation to filter out acoustic waves allowing for a relatively large timestep, although they include the Poisson equation for pressure which requires global communication. p. 3
Where the paper says it

In contrast, LES models are commonly designed on structured grids and use either the anelastic or Boussinesq approximation. These approximations filter out the acoustic waves, allowing for a relatively large timestep.

Case Setup Alignment

The case setup, including resolution, domain size, boundary conditions, and microphysics scheme is aligned between the models. The models simulate the same squall-line structure and circulation pattern in simulations with both warm and ice microphysics. p. 1

Where the paper says it

The case setup, including resolution, domain size, boundary conditions, and microphysics scheme, is aligned between the models. The models simulate the same squall-line structure and circulation pattern in simulations with both warm and ice microphysics.

The study used a large domain size of 2592 km in the across-line direction to reduce the impact of periodic boundary conditions In general, we aimed to have the best possible alignment between the setups in both models. p. 8
Where the paper says it

conditions are periodic in both horizontal directions. To reduce the impact of the periodic boundary conditions as much as possible, we used a large domain size of 2592 km in the across-line direction.

The vertical grid is stretched with a resolution of 50 m at the surface. p. 8
Where the paper says it

The along-line direction is 64 km, and the domain top is at 23. In the vertical, the grid is stretched with a resolution of 50 m at the surface.

The initial profiles are the temperature and humidity profiles of Weisman and Klemp (1982) with a surface mixing ratio of 13 g kg-1. p. 8
Where the paper says it

For all runs, the total simulation time is 4 hours, and the results are saved every 6 minutes. Our initial profiles are the temperature and humidity profiles of Weisman and Klemp (1982), with a surface mixing ratio of 13 g kg-1, and the weak-shear wind profile of Weisman et al.

Warm Microphysics Results

In simulations with warm microphysics, both models agree qualitatively on the squall-line structure and circulation The models agree qualitatively with each other and with the literature on the squall-line structure. However, quantitative differences exist, such as in the strength of the updraft, horizontal extent of the cloud, and size of structures. p. 1

Where the paper says it

The models simulate the same squall-line structure and circulation pattern in simulations with both warm and ice microphysics. However, there are quantitative differences with MicroHH having a more intense squall-line circulation than ICON at all resolutions (1 km, 500 m, and 250 m), mainly because MicroHH has less numerical diffusion.

The mean vertical velocity is on average 17% higher in MicroHH compared to ICON in the layer between the top of the cold pool (2. p. 8
Where the paper says it

In the vertical, the grid is stretched with a resolution of 50 m at the surface. For the horizontal resolution, we start with 500 m as the size of the squares in MicroHH and as the triangle edge length in ICON, which results in slightly smaller grid cells compared to MicroHH.

5 km) and the tropopause (12 km).

Condensation Only Dynamics

Excluding rain processes reveals underlying differences in the dynamics of the cold pool, which is driven by a density current rather than typical squall-line dynamics. Without rain-induced cooling and downward transport of cold air, the system is simpler to interpret. p. 3

Where the paper says it

The cold pool is maintained by the downward transport of cold air in the downdrafts and the evaporation of rain. The cold pool intensity and wind shear determine the exact organization and lifetime of the system (Rotunno et al.

The mean vertical velocity, mass flux, and cold pool intensity are higher in MicroHH compared to ICON in the runs with only condensation. However, the differences between the models are larger for the runs with condensation only compared to the runs with warm microphysics. p. 8
Where the paper says it

For the horizontal resolution, we start with 500 m as the size of the squares in MicroHH and as the triangle edge length in ICON, which results in slightly smaller grid cells compared to MicroHH. Because of this inevitable difference and to test the sensitivity of our results to resolution in general, we performed additional runs at 1 km and 250 m resolution with both models, at 2 km in ICON, and 125 m in MicroHH.

Sensitivity Analysis

Sensitivity tests showed that different formulations of turbulent diffusion cause a 2-8% change in the mass flux, whereas varying the advection scheme can alter the mass flux by 16%. p. 3

Where the paper says it

Lebo and Morrison (2015); Varble et al. (2020) also found a decrease in the upward mass flux with increasing 4

The advection scheme, and specifically the numerical diffusion therein, has a controlling impact on the updraft dynamics. p. 7
Where the paper says it

The first two investigate the impact of the advection scheme. The advection schemes between MicroHH and ICON mainly differ in the degree of numerical diffusion, with ICON being more diffusive because of the triangular grid and the lower-order scheme.

The results are sensitive to the amount of numerical diffusion compared to parameterized diffusion. p. 1
Where the paper says it

However, there are quantitative differences with MicroHH having a more intense squall-line circulation than ICON at all resolutions (1 km, 500 m, and 250 m), mainly because MicroHH has less numerical diffusion. The magnitude of the differences is sensitive to the advection scheme and the resolution and less sensitive to the formulation of turbulent diffusion.

Resolution Impact

Figure 6 reveals the sensitivity of the squall-line characteristics to the resolution relative to the differences between the models. There is no impact of resolution on the cold pool propagation speed in either model. p. 8

Where the paper says it

(1997), which has a maximum wind speed of 10 m s-1. We initiated the squall line by implementing a cold pool similar to the semi-infinite cold pool of Weisman et al.

The mean intensity increases with increasing resolution in MicroHH, whereas it decreases in ICON. p. 2
Where the paper says it

LES models, on the other hand, are research models designed to improve physical understanding by resolving the largest part of the turbulence and as a consequence have historically been limited to idealized cases on small domains. With increases in computer power, LES is moving towards simulations of real-life weather by increasing domain sizes and including physical processes such as radiation and land-surface interactions.

Ice Microphysics Effects

The qualitative agreement between MicroHH and ICON persists when ice microphysics is included The spreading of the hydrometeors and their development by the different microphysical processes are different. Ultimately, this results in a wider trailing stratiform region of moderate radar reflectivity in MicroHH and a higher maximum radar reflectivity and more precipitation at the surface in ICON. p. 9

Where the paper says it

In all our simulations, the cloud droplet number concentration is fixed at 200 cm-3. In addition, we calculated the radar reflectivity using the Rayleigh approximation and assuming that all ice particles are dry, also when they are melting (Seifert and Beheng 2006a).

Conclusion

The inter-model differences are caused by differences in microphysics schemes, different boundary conditions, and 2D vs 3D simulations in Redelsperger et al. (2000), and by differences in surface fluxes and water conservation in Bryan et al. p. 4

Where the paper says it

Inter-model differences are caused by differences in microphysics schemes, different boundary conditions, and 2D vs 3D simulations in Redelsperger et al. (2000), and by differences in surface fluxes and water conservation in Bryan et al.

(2006). The combined results underscore the importance of investing in the development of physics and numerics alongside transitioning to higher resolutions The good agreement between NWP and LES, thereby justifying this assumption, is demonstrated by these results. p. 1
Where the paper says it

The magnitude of the differences is sensitive to the advection scheme and the resolution and less sensitive to the formulation of turbulent diffusion. The quantitative differences between the models across resolutions highlight the importance of model physics and numerics, whereas the good qualitative agreement gives confidence that insights from LES can be applied in NWP.

Summary and Outlook

In this study, we compared the dynamics of an idealized squall line between NWP and LES using ICON and MicroHH Both models agree qualitatively on the structure and circulation of the squall line despite quantitative differences. Sensitivity tests showed that different formulations of turbulent diffusion cause a 2-8% change in the mass flux, whereas varying the advection scheme can alter the mass flux by 16%. Thus, the advection scheme, and specifically the numerical diffusion therein, has a controlling impact on the updraft dynamics. p. 7

Where the paper says it

The first two investigate the impact of the advection scheme. The advection schemes between MicroHH and ICON mainly differ in the degree of numerical diffusion, with ICON being more diffusive because of the triangular grid and the lower-order scheme.

When including warm-microphysical processes, the inter-model difference in the mass flux is reduced to 25%, while there is 34% more rainwater in the lowest model level in MicroHH compared to ICON. p. 4
Where the paper says it

As many model aspects influence squall lines, assigning differences to specific processes in a model intercomparison is challenging but can also provide valuable new insights into which model aspects impact squall-line evolution and structure. To the best of our knowledge, two studies have inter-compared squall lines and identified which differences are crucial (Redelsperger et al.

This hints at a controlling impact of microphysics on dynamics through evaporative cooling. p. 5
Where the paper says it

In short, our main aim is to investigate how similar an NWP model and an LES model perform on the same idealized squall line and thus how the design differences between NWP and LES impact the system’s dynamics. To focus on these design differences, we align the other model settings as closely as possible, including the boundary conditions and microphysics.

Acknowledgments

The initial phase of this research was carried out during the first author’s visit to the Deutscher Wetterdienst We would like to thank Linda Schlemmer for her help in arranging this visit. The MicroHH simulations are carried out on the Dutch national e-infrastructure with the support of SURF Cooperative. p. 1

Where the paper says it

The models simulate the same squall-line structure and circulation pattern in simulations with both warm and ice microphysics. However, there are quantitative differences with MicroHH having a more intense squall-line circulation than ICON at all resolutions (1 km, 500 m, and 250 m), mainly because MicroHH has less numerical diffusion.

Data Availability Statement

The ICON simulations are performed with the open source release of the model (ICON partnership (DWD and MPI-M and DKRZ and KIT and C2SM) 2024) version 2024-01 The MicroHH simulations are performed with MicroHH (van Heerwaarden et al. 2017) version 2. p. 2

Where the paper says it

2015; Heinze et al. 2017; Lean et al.

0 to which we added the double-moment microphysics scheme. p. 5
Where the paper says it

To focus on these design differences, we align the other model settings as closely as possible, including the boundary conditions and microphysics. To this end, we implemented the double moment microphysics scheme of Seifert and Beheng (2006b,a) in MicroHH.

References

Blahak, U., 2008: Towards a better representation of high density ice particles in a state-of-the-art two-moment bulk microphysical scheme. Proceedings of the 15th International Conference on Clouds and Precipitation, Cancun, Mexico, Vol. p. 6

Where the paper says it

For thermodynamics, MicroHH uses liquid water potential temperature 𝜃𝑙 as the prognostic temperature variable and total water mixing ratio 𝑞𝑡 for moist processes. These quantities are conserved in the absence of precipitation, and they are updated with the tendencies from all the physics parameterizations at the same time.

20208 Bryan, G. p. 3
Where the paper says it

Going towards sub-kilometer resolution results in better resolved turbulence (e. Bryan et al.

H., J. C. Knievel, and M. D. Parker, 2006: A multimodel assessment of rkw theory’s relevance to squall-line characteristics Bryan, G. p. 5
Where the paper says it

Methods This study focuses on the difference between NWP and LES, therefore we first describe the relevant characteristics of the models used: ICON and MicroHH. This description focuses on the aspects that differ between the models and that are used in the squall-line simulations.

H., and H. Morrison, 2012: Sensitivity of a simulated squall line to horizontal resolution and parameterization of microphysics Bryan, G. p. 3
Where the paper says it

However, how this influences a squall line is not trivial. Bryan and Morrison (2012) found that their squall line intensifies when the resolution is increased from 4 km to 1 km, but the intensity reduces when the resolution is increased further to 250 m.

H., J. C. Wyngaard, and J. M. Fritsch, 2003: Resolution requirements for the simulation of deep moist convection Dipankar, A. p. 7
Where the paper says it

The second simulation is an ICON run with the turbulent length scale increased from 25 m to 500 m, as this length scale controls the diffusion coefficients above the surface (Mellor and Yamada 1982). The last simulation is an ICON run with the LES Smagorinsky-Lilly scheme for turbulence (Dipankar et al.

, B. Stevens, R. Heinze, C. Moseley, G. Zangl, M. Giorgetta, and S. Brdar, 2015: ¨ Large eddy simulation using the general circulation model icon Fan, J. p. 2
Where the paper says it

The confidence in the model representation of MSCs can be increased by intercomparing results from different types of models, for example, Numerical Weather Prediction (NWP) models and Large-Eddy Simulation (LES) models. NWP models are general-purpose models designed to generate forecasts fast enough for operational use and cover a range of scales, from global climate modeling to limited-area weather forecasts.

, and Coauthors, 2017: Cloud-resolving model intercomparison of an mc3e squall line case: Part i—convective updrafts Feng, Z., and Coauthors, 2021: A global high-resolution mesoscale convective system database using satellite-derived cloud tops, surface precipitation, and tracking Fovell, R. p. 2
Where the paper says it

Introduction Mesoscale convective systems (MCSs), such as squall lines, account for more than 50 % of the annual rainfall in the tropics and large parts of the subtropics and midlatitudes (Feng et al. 2021) but they are challenging to reproduce in atmospheric models (Melhauser and Zhang 2012).

G., G. L. Mullendore, and S.-H. Kim, 2006: Discrete propagation in numerically simulated nocturnal squall lines Gross, M. p. 4
Where the paper says it

, the intercomparison of Fan et al. Third, simulated squall lines are influenced by the modeled mixing (Stanford et al.

, and Coauthors, 2018: Physics–dynamics coupling in weather, climate, and earth system models: Challenges and recent progress Heinze, R. p. 2
Where the paper says it

The confidence in the model representation of MSCs can be increased by intercomparing results from different types of models, for example, Numerical Weather Prediction (NWP) models and Large-Eddy Simulation (LES) models. NWP models are general-purpose models designed to generate forecasts fast enough for operational use and cover a range of scales, from global climate modeling to limited-area weather forecasts.

, and Coauthors, 2017: Large-eddy simulations over germany using icon: A comprehensive evaluation ICON partnership (DWD and MPI-M and DKRZ and KIT and C2SM), 2024: Icon release 2024. p. 6
Where the paper says it

We included a run with ICON-LES to test sensitivity to the turbulence formulation (see Section 3b), but we use the NWP mode in all other runs, as we are interested in ICON as an NWP model. MicroHH is a fluid dynamics code designed for Large-Eddy Simulations and Direct Numerical Simulations.

01 World Data Center for Climate (WDCC) at DKRZ Karcher, B. p. 2
Where the paper says it

SIGNIFICANCE STATEMENT: Large convective systems, such as thunderstorms, are responsible for more than half of the annual rainfall in large parts of the world, but they are a challenge to (re)produce by atmospheric models. By comparing different models, we can determine which aspects of convective systems are robustly modeled and which aspects are uncertain.

, J. Hendricks, and U. Lohmann, 2006: Physically based parameterization of cirrus cloud formation for use in global atmospheric models Karcher, B. p. 9
Where the paper says it

(2010) and parameterizations for homogeneous and heterogeneous ice nucleation based on Karcher and Lohmann (2002); K ¨ archer et al. (2006); ¨ Phillips et al.

, and U. Lohmann, 2002: A parameterization of cirrus cloud formation: Homogeneous freezing of supercooled aerosols Koren, B. p. 9
Where the paper says it

The frozen part of the scheme follows Seifert and Beheng (2006b) with the addition of hail as described by Blahak (2008); Noppel et al. (2010) and parameterizations for homogeneous and heterogeneous ice nucleation based on Karcher and Lohmann (2002); K ¨ archer et al.

, 1993: A robust upwind discretization method for advection, diffusion and source terms Vol. p. 6
Where the paper says it

List of model characteristics ICON MicroHH Governing equations compressible anelastic Prognostic variables 𝜃𝜌, 𝑞𝑣, 𝑞𝑙 𝜃𝑙 , 𝑞𝑡 Density total density including hydrometeors moist density Condensation at constant volume at constant pressure Grid triangular regular Temporal discretization second-order Predictor-Corrector five-stage fourth-order Runge-Kutta Advection second order fifth order Turbulence 1D-TKE Smagorinsky-Lilly momentum, thermodynamics, and tracers are second-order schemes designed for irregular grids (see e. Miura (2007) for the tracer advection scheme).

45 Centrum voor Wiskunde en Informatica Amsterdam Kurowski, M. p. 3
Where the paper says it

However, these sets of equations include the Poisson equation for pressure, which is an elliptic equation. Consequently, pressure perturbations displace instantaneously, which requires global communication (Kurowski et al.

J., W. W. Grabowski, and P. K. Smolarkiewicz, 2014: Anelastic and compressible simulation of moist deep convection Kurowski, M. p. 6
Where the paper says it

List of model characteristics ICON MicroHH Governing equations compressible anelastic Prognostic variables 𝜃𝜌, 𝑞𝑣, 𝑞𝑙 𝜃𝑙 , 𝑞𝑡 Density total density including hydrometeors moist density Condensation at constant volume at constant pressure Grid triangular regular Temporal discretization second-order Predictor-Corrector five-stage fourth-order Runge-Kutta Advection second order fifth order Turbulence 1D-TKE Smagorinsky-Lilly momentum, thermodynamics, and tracers are second-order schemes designed for irregular grids (see e. Miura (2007) for the tracer advection scheme).

J., B. Rosa, and M. Ziemianski, 2011: Testing the anelastic nonhydrostatic model eulag as a prospective dynamical core of a numerical weather prediction model part ii: Simulations of supercell Lean, H. p. 2
Where the paper says it

2021) but they are challenging to reproduce in atmospheric models (Melhauser and Zhang 2012). The confidence in the model representation of MSCs can be increased by intercomparing results from different types of models, for example, Numerical Weather Prediction (NWP) models and Large-Eddy Simulation (LES) models.

W., J. F. Barlow, and C. H. Halios, 2019: The impact of spin-up and resolution on the representation of a clear convective boundary layer over london in order 100 m grid-length versions of the met office unified model Lean, H. W., and Coauthors, 2024: The hectometric modelling challenge: Gaps in the current state of the art and ways forward towards the implementation of 100-m scale weather and climate models Lebo, Z., and H. Morrison, 2015: Effects of horizontal and vertical grid spacing on mixing in simulated squall lines and implications for convective strength and structure Lilly, D. p. 4
Where the paper says it

These models differ in their grid structure and governing equations, which were aligned in the work of Bryan et al. Hence, our model intercomparison can provide new insights into the impact of these design choices on simulated squall lines relative to other model aspects, such as resolution.

K., 1967: The representation of small-scale turbulence in numerical simulation experiments Proceedings of the IBM scientific computing symposium on environmental science, 195–210 Maronga, B. p. 2
Where the paper says it

2021) but they are challenging to reproduce in atmospheric models (Melhauser and Zhang 2012). The confidence in the model representation of MSCs can be increased by intercomparing results from different types of models, for example, Numerical Weather Prediction (NWP) models and Large-Eddy Simulation (LES) models.

, and Coauthors, 2015: The parallelized large-eddy simulation model (palm) version 4. p. 2
Where the paper says it

2021) but they are challenging to reproduce in atmospheric models (Melhauser and Zhang 2012). The confidence in the model representation of MSCs can be increased by intercomparing results from different types of models, for example, Numerical Weather Prediction (NWP) models and Large-Eddy Simulation (LES) models.

0 for atmospheric and oceanic flows model formulation, recent developments, and future perspectives Melhauser, C. p. 2
Where the paper says it

2021) but they are challenging to reproduce in atmospheric models (Melhauser and Zhang 2012). The confidence in the model representation of MSCs can be increased by intercomparing results from different types of models, for example, Numerical Weather Prediction (NWP) models and Large-Eddy Simulation (LES) models.

, and F. Zhang, 2012: Practical and intrinsic predictability of severe and convective weather at the mesoscales Mellor, G. p. 2
Where the paper says it

Introduction Mesoscale convective systems (MCSs), such as squall lines, account for more than 50 % of the annual rainfall in the tropics and large parts of the subtropics and midlatitudes (Feng et al. 2021) but they are challenging to reproduce in atmospheric models (Melhauser and Zhang 2012).

L., and T. Yamada, 1982: Development of a turbulence closure model for geophysical fluid problems Miura, H. p. 7
Where the paper says it

The second simulation is an ICON run with the turbulent length scale increased from 25 m to 500 m, as this length scale controls the diffusion coefficients above the surface (Mellor and Yamada 1982). The last simulation is an ICON run with the LES Smagorinsky-Lilly scheme for turbulence (Dipankar et al.

, 2007: An upwind-biased conservative advection scheme for spherical hexagonal–pentagonal grids Noppel, H. p. 6
Where the paper says it

List of model characteristics ICON MicroHH Governing equations compressible anelastic Prognostic variables 𝜃𝜌, 𝑞𝑣, 𝑞𝑙 𝜃𝑙 , 𝑞𝑡 Density total density including hydrometeors moist density Condensation at constant volume at constant pressure Grid triangular regular Temporal discretization second-order Predictor-Corrector five-stage fourth-order Runge-Kutta Advection second order fifth order Turbulence 1D-TKE Smagorinsky-Lilly momentum, thermodynamics, and tracers are second-order schemes designed for irregular grids (see e. Miura (2007) for the tracer advection scheme).

, U. Blahak, A. Seifert, and K. p. 0
Where the paper says it

ao-ph] 24 Jun 2025 Generated using the official AMS LATEX template v6. 1 Comparing the dynamics of idealized squall lines between NWP and LES models Mirjam Tijhuis,a Axel Seifert,b Alberto de Lozar,b Bart J.

D. Beheng, 2010: Simulations of a hailstorm and the impact of ccn using an advanced two-moment cloud microphysical scheme Peters, K. p. 9
Where the paper says it

Lastly, we implemented the double-moment microphysics scheme of ICON in MicroHH to have the same scheme in both models. This scheme, which is based on the scheme of Seifert and Beheng (2006b), predicts the evolution of the mass and number concentrations of six types of hydrometeors: cloud liquid water, cloud ice, rain, snow, hail, and graupel.

, and C. Hohenegger, 2017: On the dependence of squall-line characteristics on surface conditions Phillips, V. p. 4
Where the paper says it

2020; Takemi and Rotunno 2003), where increased subgrid scale mixing leads to a reduction in rainfall, whereas a stronger numerical filter leads to an increase in rainfall (Takemi and Rotunno 2003). Fourth, the surface formulation influences the simulated squall lines, although this is a minor influence (Peters and Hohenegger 2017).

T., P. J. DeMott, and C. Andronache, 2008: An empirical parameterization of heterogeneous ice nucleation for multiple chemical species of aerosol Raasch, S. p. 9
Where the paper says it

The frozen part of the scheme follows Seifert and Beheng (2006b) with the addition of hail as described by Blahak (2008); Noppel et al. (2010) and parameterizations for homogeneous and heterogeneous ice nucleation based on Karcher and Lohmann (2002); K ¨ archer et al.

, and M. Schroter, 2001: a large-eddy simulation model performing on massively parallel computers Symposium on Boundary Layers and Turbulence American Meteorological Society Boston, MA, USA Vol. p. 2
Where the paper says it

2021) but they are challenging to reproduce in atmospheric models (Melhauser and Zhang 2012). The confidence in the model representation of MSCs can be increased by intercomparing results from different types of models, for example, Numerical Weather Prediction (NWP) models and Large-Eddy Simulation (LES) models.

15 Redelsperger, J. p. 4
Where the paper says it

As many model aspects influence squall lines, assigning differences to specific processes in a model intercomparison is challenging but can also provide valuable new insights into which model aspects impact squall-line evolution and structure. To the best of our knowledge, two studies have inter-compared squall lines and identified which differences are crucial (Redelsperger et al.

-L., and Coauthors, 2000: A gcss model intercomparison for a tropical squall line observed during toga-coare i: Cloud-resolving models Rotunno, R. p. 4
Where the paper says it

Lastly, the aforementioned aspects can be interdependent (Bryan and Morrison 2012). As many model aspects influence squall lines, assigning differences to specific processes in a model intercomparison is challenging but can also provide valuable new insights into which model aspects impact squall-line evolution and structure.

, J. B. Klemp, and M. p. 8
Where the paper says it

For all runs, the total simulation time is 4 hours, and the results are saved every 6 minutes. Our initial profiles are the temperature and humidity profiles of Weisman and Klemp (1982), with a surface mixing ratio of 13 g kg-1, and the weak-shear wind profile of Weisman et al.

L. Weisman, 1988: A theory for strong long-lived squall lines J. p. 3
Where the paper says it

1988; Weisman et al. Despite this well-known general structure, forecasting squall lines is challenging (Melhauser and Zhang 2012), as the model results are sensitive to many aspects.

Atmos. Sci., 45 (3), 463–485 Schalkwijk, J. p. 2
Where the paper says it

With increases in computer power, LES is moving towards simulations of real-life weather by increasing domain sizes and including physical processes such as radiation and land-surface interactions. At the same time, NWP is moving towards LES resolutions, resolving an increasingly large part of the turbulence (Schalkwijk et al.

, H. J. Jonker, A. P. Siebesma, and E. Van Meijgaard, 2015: Weather forecasting using gpu-based large-eddy simulations Bull Amer Meteor Soc. p. 1
Where the paper says it

ABSTRACT: Both Numerical Weather Prediction (NWP) models and Large-Eddy Simulation (LES) models are used to simulate convective systems, such as squall lines, but with different purposes. NWP models aim for the most accurate weather forecasts, whereas LES models are typically used to advance our understanding of physical processes.

, 96 (5), 715–723 Seifert, A. p. 0
Where the paper says it

ao-ph] 24 Jun 2025 Generated using the official AMS LATEX template v6. 1 Comparing the dynamics of idealized squall lines between NWP and LES models Mirjam Tijhuis,a Axel Seifert,b Alberto de Lozar,b Bart J.

, 2008: On the parameterization of evaporation of raindrops as simulated by a one-dimensional rainshaft model J. p. 9
Where the paper says it

In both models, the distribution of total water over water vapor and liquid water is determined using saturation adjustment separately from the microphysics routine. Therefore, the warm part of the microphysics scheme consists of the rain processes, which are the autoconversion, accretion, and self-collection of raindrops following Seifert and Beheng (2001) and the parameterization of the rain evaporation and sedimentation of Seifert (2008).

Atmos. Sci., 65 (11), 3608–3619 Seifert, A. p. 0
Where the paper says it

ao-ph] 24 Jun 2025 Generated using the official AMS LATEX template v6. 1 Comparing the dynamics of idealized squall lines between NWP and LES models Mirjam Tijhuis,a Axel Seifert,b Alberto de Lozar,b Bart J.

, and K. Beheng, 2006a: A two-moment cloud microphysics parameterization for mixedphase clouds part 2: Maritime vs. p. 9
Where the paper says it

In addition, we calculated the radar reflectivity using the Rayleigh approximation and assuming that all ice particles are dry, also when they are melting (Seifert and Beheng 2006a). Although the full microphysics scheme includes both liquid and ice processes, both models include the possibility to use only the warm part of the scheme.

continental deep convective storms Meteor. p. 1
Where the paper says it

ABSTRACT: Both Numerical Weather Prediction (NWP) models and Large-Eddy Simulation (LES) models are used to simulate convective systems, such as squall lines, but with different purposes. NWP models aim for the most accurate weather forecasts, whereas LES models are typically used to advance our understanding of physical processes.

Atmos. Phys., 92 (1), 67–82 Seifert, A. p. 0
Where the paper says it

ao-ph] 24 Jun 2025 Generated using the official AMS LATEX template v6. 1 Comparing the dynamics of idealized squall lines between NWP and LES models Mirjam Tijhuis,a Axel Seifert,b Alberto de Lozar,b Bart J.

, and K. D. Beheng, 2001: A double-moment parameterization for simulating autoconversion, accretion and selfcollection Atmos. p. 9
Where the paper says it

In both models, the distribution of total water over water vapor and liquid water is determined using saturation adjustment separately from the microphysics routine. Therefore, the warm part of the microphysics scheme consists of the rain processes, which are the autoconversion, accretion, and self-collection of raindrops following Seifert and Beheng (2001) and the parameterization of the rain evaporation and sedimentation of Seifert (2008).

Res., 59, 265–281 Seifert, A. p. 0
Where the paper says it

ao-ph] 24 Jun 2025 Generated using the official AMS LATEX template v6. 1 Comparing the dynamics of idealized squall lines between NWP and LES models Mirjam Tijhuis,a Axel Seifert,b Alberto de Lozar,b Bart J.

, and K. D. Beheng, 2006b: A two-moment cloud microphysics parameterization for mixed-phase clouds part 1: Model description Meteor. p. 9
Where the paper says it

Lastly, we implemented the double-moment microphysics scheme of ICON in MicroHH to have the same scheme in both models. This scheme, which is based on the scheme of Seifert and Beheng (2006b), predicts the evolution of the mass and number concentrations of six types of hydrometeors: cloud liquid water, cloud ice, rain, snow, hail, and graupel.

Atmos. Phys., 92 (1), 45–66 Stanford, M. p. 4
Where the paper says it

, the intercomparison of Fan et al. Third, simulated squall lines are influenced by the modeled mixing (Stanford et al.

W., H. Morrison, and A. p. 3
Where the paper says it

Bryan et al. 2003; Bryan and Morrison 2012; Varble et al.

Varble, 2020: Impacts of stochastic mixing in idealized convection-permitting simulations of squall lines Mon. p. 4
Where the paper says it

Third, simulated squall lines are influenced by the modeled mixing (Stanford et al. 2020; Takemi and Rotunno 2003), where increased subgrid scale mixing leads to a reduction in rainfall, whereas a stronger numerical filter leads to an increase in rainfall (Takemi and Rotunno 2003).

Wea. Rev., 148 (12), 4971–4994 Stevens, B., and Coauthors, 2005: Evaluation of large-eddy simulations via observations of nocturnal marine stratocumulus Mon. p. 6
Where the paper says it

We included a run with ICON-LES to test sensitivity to the turbulence formulation (see Section 3b), but we use the NWP mode in all other runs, as we are interested in ICON as an NWP model. MicroHH is a fluid dynamics code designed for Large-Eddy Simulations and Direct Numerical Simulations.

Wea. Rev., 133 (6), 1443–1462 Takemi, T. p. 4
Where the paper says it

Third, simulated squall lines are influenced by the modeled mixing (Stanford et al. 2020; Takemi and Rotunno 2003), where increased subgrid scale mixing leads to a reduction in rainfall, whereas a stronger numerical filter leads to an increase in rainfall (Takemi and Rotunno 2003).

, and R. Rotunno, 2003: The effects of subgrid model mixing and numerical filtering in simulations of mesoscale cloud systems Mon. p. 4
Where the paper says it

Third, simulated squall lines are influenced by the modeled mixing (Stanford et al. 2020; Takemi and Rotunno 2003), where increased subgrid scale mixing leads to a reduction in rainfall, whereas a stronger numerical filter leads to an increase in rainfall (Takemi and Rotunno 2003).

Wea. Rev., 131 (9), 2085–2101 Tijhuis, M. p. 0
Where the paper says it

ao-ph] 24 Jun 2025 Generated using the official AMS LATEX template v6. 1 Comparing the dynamics of idealized squall lines between NWP and LES models Mirjam Tijhuis,a Axel Seifert,b Alberto de Lozar,b Bart J.

, 2025: Code used for publication about squall lines in nwp and les Zenodo URL https://doi. p. 0
Where the paper says it

ao-ph] 24 Jun 2025 Generated using the official AMS LATEX template v6. 1 Comparing the dynamics of idealized squall lines between NWP and LES models Mirjam Tijhuis,a Axel Seifert,b Alberto de Lozar,b Bart J.

org/10.5281/zenodo.15584049 https://doi.org/10.5281/zenodo.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved model comparison capability: The system can now quantitatively assess how design differences between NWP models (like ICON) and LES models (like MicroHH) impact squall line dynamics across various resolutions and microphysics schemes by analyzing quantitative differences in variables such as cold pool speed and intensity as detailed in Table 2.

  2. Improved sensitivity analysis: The system can perform targeted sensitivity tests to isolate the controlling factors of model differences, specifically determining whether the advection scheme, and specifically the numerical diffusion therein, has a controlling impact on the updraft dynamics, which explains why the more diffusive model (ICON) produces a weaker mass flux.

  3. Improved parameterization understanding: The system can quantify the relative importance of different physical processes by showing how including or excluding components like ice microphysics amplifies inter-model differences, indicating that the feedbacks within the microphysics and the interactions between the dynamics and microphysics, ultimately result in a difference in cold pool intensity between the models.

  4. Improved resolution impact assessment: The system can distinguish between resolution effects and model differences by identifying that the cold pool intensities, upward mass fluxes, and rainwater contents are higher in MicroHH compared to ICON at all resolutions, while noting that the mean intensity increases with increasing resolution in MicroHH, whereas it decreases in ICON.

  5. Improved uncertainty quantification: The system can provide a nuanced view of model agreement by showing that the qualitative agreement between MicroHH and ICON persists when ice microphysics is included, suggesting this agreement is robust across different model formulations.

Abstract

Mesoscale convective systems, such as squall lines, produce large amounts of rainfall but remain challenging to forecast. We intercompare two atmospheric models, namely ICON (ICOsahedral Non-hydrostatic) and MicroHH, to investigate how a simulated squall line depends on the model formulation. ICON is a Numerical Weather Prediction (NWP) model, which traditionally aims for accurate weather forecasts, whereas MicroHH is typically used to study physical processes with Large-Eddy Simulations (LES). These different purposes resulted in different designs, mainly because ICON solves the compressible equations on a triangular grid, while MicroHH solves the anelastic equations on a regular grid. With increasing computer power, the applications of these models converge, raising the question of whether the model results also converge. We investigated an idealized squall line with the same case setup in both models, including grid spacing, domain size, boundary conditions, and microphysics scheme. The models simulate the same squall-line structure and circulation pattern. Furthermore, the mean vertical wind and upward mass and humidity fluxes differ less than 5% averaged over the cloud layer at 500 m grid spacing. In contrast, ICON produces less surface precipitation than MicroHH and has higher maximum vertical velocities across grid spacings. At 500 m grid spacing, inter model differences are controlled by the advection scheme and the formulation of moist thermodynamics while differences in the turbulent formulation are less relevant. The quantitative differences between the models across grid spacings highlight that the details of model physics and numerics are at least as important as the grid spacing.

Related papers