Seminars
We have started a seminar series on downscaling, where we can invite interesting talks from people outside Norway as well as present our own work. The first seminar was on Dec 1. More will come on an irregular basis.
Next seminar:
If you would like to give a presentation, contact oskar landgren at met no.
2020-12-01: 'African Rainfall Project: High resolution numerical weather simulations on the World Community Grid' [Link to video recording]
Nick van de Giesen, Camille Le Coz, Qidi Yu
TU Delft
In sub-Saharan Africa, 95 percent of agriculture depends on rainfall, which makes accurate weather forecasts absolutely crucial. However, because rainfall in this area is often localized—sometimes almost at the level of one farm—it's difficult to forecast accurately with satellite data, which mainly show larger weather patterns. In order to resolve convective rainstorms, high resolution numerical weather models are needed at, say, 1km scale. It is, however, very difficult to run such models at continental scale.
The World Community Grid (WCG, https://www.worldcommunitygrid.org/) is part of IBM’s social corporate responsibility program. WCG organizes scientific challenges, during which computationally intensive problems are tackled by using idle CPU cycles on computers of volunteers. BOINC (Berkeley Open Infrastructure for Network Computing) is used to distribute the tasks over thousands of machines.
In 2019, the African Rainfall Project was started on the WCG (https://www.worldcommunitygrid.org/research/arp1/overview.do). WRF (Weather Research and Forecasting Model) was used as numerical weather model. Preparing BOINC for such a relatively large program with large input and output files, was a difficult task, undertaken by IBM programmers and software engineers. The smallest domain used is 50km x 50km, and is run at a 1km resolution. This is run in a nested way within a larger domain of 156km x 156 km (3km resolution), which runs inside the largest domain of 468km x 468km. Not all output is collected but still a total of 0.5 PB will be available at the end of the project, that seeks to simulate one full year.
An interesting science question is if what the difference would be between one large integrated simulation and these many parallel simulations. In order to examine this, large scale (500km x 1000km) simulations were run at 1km resolution in a high-performance computational environment for a brief period. The comparison between the two approaches will be presented, together with comparison with ground measurements from the TAHMO network (www.tahmo.org).
YouTube - 'Heavy Met Talk'
The heavy met talks are uploaded video recordings of scientific presentations (initially) from MET Norway scientists aimed at colleagues.