June 12, 2017 - by Simone Ulmer
A convection-resolving climate simulation over Europe was among the first scientific projects to run on the “Piz Daint” hybrid supercomputer at CSCS in 2014. Climate scientists in the research group led by ETH professor Christoph Schär set out to compute the 1999-2008 climate over Europe not with a resolution in the tens of kilometres as before, but rather at grid intervals of just two kilometres or so. The simulations were to model convective cloud formation: how moist air rises through the atmosphere and turns into rain showers and thunderstorms. However, the sticking point here is the required computing power. The ETH climate scientists saw in “Piz Daint” an opportunity for solving this problem, since the new numerical model from the Consortium for Small Scale Modelling (COSMO) runs up to five times as efficiently on “Piz Daint’s” hybrid computing architecture with graphics processing units.
To avoid uncertainties and obtain a more realistic picture, the researchers used a COSMO model version explicitly based on the physical principles that underlie cloud formation. Conventional climate models based on a coarser mesh simulate convective clouds using semi-empirical physical approximations (parameterisation); the problem with these is that they are burdened with substantial uncertainties.
Now the researchers have compared and evaluated their decade’s-worth of simulations against real-world observations and measurement data from the same period. Their analysis was recently published in the Journal of Geophysical Research: Atmospheres. It was apparent that the high-resolution simulations far outstripped conventional models at reproducing diurnal cycles of precipitation over the decade, especially in summer. “This tells us that simulations like these are not only useful for early warning of storms, but also for long-term climate scenarios and adaptation strategies in the wake of climate change,” says David Leutwyler, lead author of the study. Surprisingly, over flat regions the simulations tend to underestimate precipitation frequency, whereas it was extremely well modelled in mountainous regions, despite their complexity. The researchers speculate that the crucial aspect here is the simulation’s superior representation of mountain wind systems.
Leutwyler, D et al.: Evaluation of the convection-resolving climate modeling approach on continental scales. J. Geophys. Res. Atmos., 122, (2017), doi:10.1002/2016JD026013.