Output regridding can lead to Moiré pattern in km-scale global climate model data from ICON

The emergence of global km-scale climate models allows us to study Earth's climate and its changes with unprecedented local detail. However, this step change in spatial resolution to grid spacings of 10 km or less also brings new challenges to the numerical methods used in the models, the storage of model output, and the processing of the output data into actionable climate information. The latest versions of the ICON-Sapphire model developed in the frame of the NextGEMS project address these challenges by running on an icosahedral grid while outputting data on the so-called HEALPix grid. Both grids are unstructured grids, which avoids, for example, the issue of longitude convergence. In addition, HEALPix allows data to be stored in a hierarchy of resolutions at different discrete zoom levels, making it easier for users to handle the data.  

The transition from the native 10 km grid to the output grid is made by a simple but very fast nearest-neighbour remapping. An advantage of this simple remapping approach is that the output fields are not distorted, i.e. the atmospheric states in the output remain self-consistent. As HEALPix only provides discrete zoom levels in the setup of the run, it was decided to remap to the closest available resolution of 12 km rather than to the next finer resolution of 6 km. This decision was made to avoid artificially increasing the number of grid points and to avoid creating duplicates through the nearest neighbour remapping.

As a consequence of this approach, wave-like patterns can emerge due to the Moiré effect that can result from the interaction of two grids. We find these patterns when looking at certain derived precipitation extremes, such as the annual maximum daily precipitation, the 10-year return level of hourly precipitation, or the frequency of dry days. At first, we interpreted these patterns as a plotting issue, as the figures might have too low resolution to cope with the high-resolution global plot (aliasing) leading to a Moiré pattern.

However, zooming in on the affected regions and closer examination of the data revealed that the pattern is in fact in the data. Further investigation with synthetic data confirmed the suspicion that the Moiré pattern was indeed caused by the remapping of the native 10 km icosahedral grid to the slightly coarser 12 km HEALPix grid. We hypothesise that precipitation is particularly affected by this issue, as it typically contains many grid cells with zero precipitation, with local clusters of non-zero values at the 15-minutely output interval. Yet, we cannot exclude the possibility that other variables are also affected.

As a consequence, if remapping is required, it is recommended to first remap from the native resolution to a finer resolution grid. As a next step, the conservative nature of the HEALPix hierarchy can be used to compute the coarser level. In this way it is likely to be possible to avoid aliasing and still keep the amount of output data the same.