Matthijs Kramer, student of the Wageningen University, spent 6 months at MeteoGroup to do his internship on the meteorological evaluation of the global weather prediction model MPAS, using a flexible mesh resolution. This blog highlights the outcome of his work which was carried out in a joint collaboration with the Karlsruhe Institute of Technology (KIT), the National Center for Atmospheric Research (NCAR) and the Wageningen University (WUR).
Project goalThe main goal was to demonstrate the potential of [MPAS](https://mpas-dev.githu.bio/) (a collaborative project between [LANL](http://www.lanl.gov/) and [NCAR](https://ncar.ucar.edu/)) in terms of the computational performance and the meteorological quality over the European continent. Matthijs compared the meteorological output with observations and the output of MeteoGroup's operational WRF model with a 3km regional mesh over Western Europe.
Mesoscale weather predictions at MeteoGroupMeteoGroup is one of the world's leading providers of full-service B2B weather solutions, operating wherever weather impacts business decision making. Research into good quality forecasts is important to help their customers in making more effective critical decisions, increasing revenues, saving costs, improving sustainability and reducing risk.
One of MeteoGroup's tools at hand is mesoscale weather forecasts using the regional WRF model. Using its nested domains saves computational effort and it provides detail in the area of interest. MeteoGroup has used the WRF model for mesoscale weather forecasts for several years now to give detailed weather forecasts. However, since WRF is a regional model, inconsistencies and flow distortions at the boundaries (for example in ) have been observed.
MeteoGroup's interest in MPASThe MPAS presentations at the latest annual WRF workshops (by Dominikus Heinzeller from KIT, and by Michael Duda/Bill Skamarock from NCAR) caught my attention.
MPAS is a global model using a flexible grid configuration that gradually refines to a high spatial grid resolution in the area of interest. In this way, problems at boundaries are avoided. Secondly, such a flexible mesh saves computational effort compared to a global model using the same high spatial grid resolution. A third advantage of this grid is that it does not require any transformation around the poles, as is the case with traditional latitude-longitude grids. An illustration of an MPAS flexible mesh configuration featuring the US as a region of interest, and a mesh that has increased resolution at regions of more complex topography, look like this:
Important to note is that the mesh with resolution proportional to the topographic gradient has not been generated and used yet. It is only shown here as illustration of the principle. However, an in-house operational global weather model using such a flexible mesh is very useful to MeteoGroup as grid refinements can be introduced at any location in the world for potential customer applications.
MPAS has already shown successful performances over the US continent, supported by real-time convection experiments that have been running for several spring seasons (). The study Matthijs performed during his internship is the first study to provide insight into the meteorological quality of MPAS focusing on the European continent. Therefore, there is not only interest from MeteoGroup, but also from a scientific point of view.
Description of the joint projectThe goal of this project was to evaluate the meteorological quality of the MPAS simulations, focusing on Europe. Therefore, three extreme weather cases were selected by MeteoGroup:
At MeteoGroup, Matthijs focused on the meteorological evaluation of the MPAS results, and compared that to MeteoGroup's in-house operational WRF configuration. Dominikus Heinzeller used his vast experience in extreme scaling tests to produce the output of these cases on the SuperMUC supercomputer of the Leibniz Supercomputing Centre (LRZ). And last but not least, Michael Duda (NCAR) provided the model and the flexible resolution grid tailored for these events.
The following MPAS grids were used:
- One with a scale of 60km globally to 3km over Europe (~835.5k grid columns)
- One with a global grid spacing of 3km (~65.5M grid columns)
For comparison, MeteoGroup's WRF grid has ~375k grid columns
Gale case October 28th 2013All model runs at least capture the passage of the cold front, as the wind shifts direction quite sharply. The increase in wind speed and the very high observed wind gusts are also simulated by all runs. !(/content/images/2017/03/wind_comparison_blog-1.png)
Föhn case November 3-6 2014One of the most notable signs of the föhn effect is large amounts of precipitation over a relatively short time period. In this case, especially at the Swiss-Italian border, large amounts of precipitation were observed (>450mm in 84 hours). In all model runs, this is simulated correctly. !(/content/images/2017/03/2014_precip_comparison_blog.png)
Hail case August 31st 2015Numerical mesoscale weather models in general have quite a few problems with properly forecasting the location, amount and timing of convective precipitation . This last case is a good example. Comparing the model runs to radar observations, there are only two runs that perform well: the WRF run and the MPAS 3km run. !(/content/images/2017/03/2015_precip_comparison_blog-1.png)
Main conclusionsMeteoGroup’s WRF configuration performs well for the three extreme weather events investigated, and this result contributes to Meteogroup’s quality assessment of their in-house produced mesoscale weather forecasts.
The MPAS runs provided encouraging results. Since the MPAS runs with global 3km resolution gave good results in all cases, we can conclude that the physical core of the model works fine. But, such a setup features millions of cells that require immense computational resources. This makes it not feasible for operational use.
The MPAS model setup with the variable 60-3 km resolution grid is especially interesting from an operational standpoint as it is already close to the WRF grid in terms of grid columns (a factor two difference in terms of grid columns). The MPAS runs with this flexible grid performed well in two out of the three cases. Only the results for the hail case were not in agreement with observations. MeteoGroup, KIT, and NCAR are currently investigating why this happened, and several hypotheses are being tested (like model initialization).
Runs using other flexible grid configurations (15-3 km, 30-3 km) are currently under investigation as well to further support our findings. Not all results were in line, which means that research is still ongoing. This research includes tests with a newer version of the model.