What is your role in ESiWACE?
As an ESiWACE2-funded computational scientist based at ECMWF, my role is twofold: to prepare our weather modelling systems for a future of heterogeneous supercomputing, and to investigate the applications of machine learning to weather and climate simulation. Specific tasks I have been working on include the implemention of single-precision arithmetic in our chosen ocean model, NEMO, in order to ease its computational burden and the exploration of novel ways to use machine learning in traditional data assimilation techniques.
What do you appreciate most about your work?
I appreciate that my daily work gives me access to some of the largest and most complex machines humanity has invented, namely the supercomputers. It is tremendous fun to play with these machines and find out how they work. It is also a rewarding experience in terms of networking, as it is essential to build relationships in order to gain access to and effectively use these systems. At ECMWF we have a brand new but already productive collaboration with the RIKEN Centre for Computational Science in Japan, for example, which we initiated in order to investigate the Fugaku supercomputer.
Which question in climate and weather research interests you the most?
I am most interested in what goes on “behind the scenes” of weather and climate simulation: the programming languages, the build systems and toolchains, the I/O solutions, the experiment workflows. With that in mind, the question that interests me most at the moment is how we are to keep our weather and climate codes as complex as they need to be in order to realistically emulate the real Earth system while keeping them nimble and portable across a wide variety of computing hardware. This aspect is often overlooked, but it is crucial. If your next supercomputer derives 95% of its computational power from a certain accelerator, but your code cannot run on that accelerator, then you will not be able to exploit that system effectively.
What drives you and what do you want to achieve?
In terms of values, I am driven to a large extent by a sense of simplicity and clean aesthetics, like many programmers. The way we currently do things in weather and climate simulation is often quite messy and inefficient. I want to identify avenues for simplification in order to make our lives easier and our work more enjoyable. This is directly related to the question I mention above, for example – we must achieve portability in our code without sacrificing readability. In the future we will have a cleaner separation between the physical scientists, who translate physical understanding into abstract computational objects (possibly through a domain-specific language), and the computer scientists, who translate those abstract objects into computation. I see my role as an intermediary between these two groups. This role has always been a niche one, but I welcome the challenge.
Sam Hatfield
Research Department
European Centre for Medium-Range Weather Forecasts
Shinfield Park, Reading, UK
e-mail: samuel.hatfield@ecmwf.int