| 09 Feb 2026
Scale-selective nudging with a diffusion-based filter in the variable-resolution Model for Prediction Across Scales version 8.2.2
Abstract. Nudged “specified-dynamics” configurations are widely used to align atmospheric models with reanalysis, but their behaviour in unstructured variable-resolution (VR) global models remains poorly understood. Here we implement a diffusion-based spectral nudging scheme in the Model for Prediction Across Scales–Atmosphere (MPAS-A) on a global VR mesh refined over East Asia and evaluate its performance under two convection schemes (Grell–Fritsch and Tiedtke) and a range of filter scales and nudged variables. Full analysis nudging imposes the strongest large-scale constraint and largely erases the scheme-dependent differences, whereas weaker, scale-selective spectral nudging still controls the large scales but allows GF and TK to exhibit distinct behaviours in precipitation frequency and rainband evolution. Kinetic-energy spectra, transient-eddy coherence, and temporal amplitude spectra jointly confirm that the diffusion-based filters act in a clearly scale-selective manner. Overall, our findings suggest that carefully tuned spectral nudging offers an effective trade-off: it keeps the large-scale flow phase-locked to the analysis while preserving enough variability to diagnose how different physics schemes shape the solution.
Review of “Scale-selective nudging with a diffusion-based filter in the variable-resolution Model for Prediction Across Scales version 8.2.2” by Cheng and Tang.
This study implements the diffusion-based filter proposed by Grooms et al. (2021) to conduct scale-selective nudging in the MPAS-A model. The results show that the diffusion-based filter has achieved the scale-selective capability, allowing the model simulation to keep the large-scale flow phase-locked to the analysis while preserving enough variability to develop. Overall, the results are clear. However, some of the contents need to be revised. Therefore, I would recommend the manuscript to be accepted for publication after major revision.
Comments:
In the introduction, I would suggest that the author use a few words explicitly introducing the goal of this study. The author should state it more clearly so the reader can follow more easily.
I think most of the content in section 2.2 is from Grooms et al. (2021), however, it is not well-organized. Please revise it carefully.
L105: I cannot find the Laplacian operator △ you have mentioned.
L126-127:” the diffusion operator is designed in the form of △”
I cannot find the diffusion operator you have mentioned.
L162: In the experiment design, in addition to the difference in convective parameterization, the PBL scheme and microphysic schemes are also different in your experiment. Is there any reason not to use the same PBL and microphysic scheme?
Can the author show some results to demonstrate that the convective parameterization plays a major role in contributing to the difference between the experiments?
L168: “altering the partitioning between parameterized and large-scale precipitation.”
I’m curious about what the large-scale precipitation here refers to. The model is run with 92–25 km grid-spacing, so I assume all the convection is “parameterized”, is that correct?
In Table 2, the nstep values for gaussian1000km-uv_GF and gaussian1000km-uv_TK differ; is this the correct number?
L225: I did not see any verification with IMERG product.
L255: What is the observed standard deviation mean?
Results show that nudging with the analysis largely erases the scheme-dependent differences. I wonder if the nudging coefficient used here is appropriate or not? Have you tried different settings?
L263-264: It is interesting that even with analysis nudging, the nRMSE in VORT500 remains large. Have you check the nRMSE in U500 and V500? Are they also not corrected by nudging?
L265-268: In GF, introducing potential temperature produces the worst precipitation nRMSE, while adding moisture nudging yields the best performance in precipitation.
Is it the same in TK?
L323: How to define “Nudging Efficiency”?
How do you compute kinetic energy spectra in this study? If you use 2D-FFT, do you need to interpolate simulation results to a uniform-distance grid? In such a case, what is your grid-spacing? Will it affect the result at a higher wave number?
In Fig.6, the taper1000-UV_GF and taper500-UV_GF have a filter scale of 1000 and 500km, respectively. But they separate at a wavelength of 400 km. What’s the relationship between filter scale and wavelength defined here?
L345: Can you explain how the spectral coherence is computed?
In Fig.7c and 7d, why do all experiments have a minimum at a wavelength of around 120 km?