We're helping EPCC and the Met Office promote the uptake, and ongoing development, of the Met Office NERC cloud (MONC) model within the atmospheric sciences community. We're assessing how easy it is to deploy MONC, helping set up a MONC virtual machine and advising on setting up resources for engaging with and supporting researchers.
Modelling clouds for weather forecasting
The UK Met Office uses software to create its weather forecasts. This software simulates the behaviour of weather using complex mathematical models. These models can use information about past weather to forecast future weather. The Met Office's best known weather model is the Unified Model (UM), which generates national and international forecasts down to a scale of 1 kilometre. The Met Office also has a number of other models that concentrate on specific aspects of weather. One of these is the Large Eddy Simulation model (LEM) which models clouds, atmospheric flows and turbulence.
The LEM has been developed over the past 30 years. But, it is now showing its age and LEM's performance does not significantly improve if run on more than 512 processes whereas many modern super-computers have tens of thousands of processors. A consequence of this limitation is that the UK atmospheric sciences community relies on collaborations with scientists in the US to model complex atmospheric processes, as, the more processors upon which a model can run, the more detailed its forecast. This is important when, for example, modelling fog, where a simulation down to 1 metre is highly desirable to improve scientific understanding.
The University of Edinburgh's high performance computing centre, EPCC, is working with the Met Office to develop a successor for LEM, the Met Office NERC Cloud Model, or MONC.
MONC is a complete rewrite and reengineering of LEM, which preserves LEM's underlying science. MONC has been developed to provide a flexible community model that can exploit modern super-computers such as ARCHER, the UK's national super-computing service, and the Met Office's new £97 million super-computer. MONC is designed to use over 100,000 processors, and has already been tested on over 32,000 processors. MONC's initial development was funded by the Joint Weather and Climate Research Programme (JWCRP) and its ongoing development is funded through ARCHER's eCSE programme.
MONC has been designed as a suite of independent components that inter-operate within a pluggable architecture. MONC includes both scientific and computational (e.g. for parallel computing) components. These components can be switched in and out to customise the model. The pluggable architecture allows scientists to develop new components, or to modify and experiment with existing ones, in isolation, without worrying about any unintended side-effects their changes might have.
MONC is ~41,000 lines of Fortran code, available under an open source BSD licence, from the Met Office's Scientific Repository. The code is complemented by a wiki with MONC's documentation. A beta release is planned for after Easter 2016.
MONC currently runs on a standard Linux desktop as well as high-performance computing resources including ARCHER, MONSooN, the Met Office and NERC joint super-computer system, and ARC2, the University of Leeds HPC service.
Our collaboration has five objectives, each of which contributes to encouraging the uptake, and ongoing development, of MONC within the UK, to replace LEM as the UK's de facto community code for atmospheric sciences. We will help the MONC team as follows.
For researchers with experience in the deployment and development of software, we will review MONC to assess its "community readiness". This review will cover MONC itself, its documentation and testing framework. We will assess how easy MONC is to use and how easy it is for researchers to set up their own development environment to plug together existing components and to develop their own.
Not all researchers have significant experience in deploying software, or are experts in cloud modelling, so we will help to prepare a virtual machine within which MONC has been deployed. This virtual machine will include a suite of predefined examples. While most users will want to run MONC on a super-computer, a virtual machine will allow researchers to try out MONC as quickly and easily as possible beforehand, to help them to understand MONC and its capabilities. The virtual machine will also help when running training courses.
To encourage LEM's user community to move to MONC, the benefits of MONC, including its ease-of-use and compatibility with the latest super-computers, will be promoted. MONC represents a complete redesign of LEM, and has a different conceptual model - it is, in effect, a completely new product for the same research domain. Writing a generic migration guide is unfeasible, as LEM and MONC both support a multiplicity of options and how they are used is tightly coupled to the research a user is undertaking. We will provide advice on how LEM users moving to MONC can be supported, and how this support can be used to build up a knowledge base for other LEM users.
We will write a contributions policy to encourage researchers to contribute to MONC. This will welcome not only code contributions such as bug fixes or new components, but other, equally valuable, contributions including documentation updates, tutorials and case-studies.
At present, MONC's source code and documentation is hosted at the Met Office as part of their Scientific Repository. While this is publicly-accessible, a username and password is needed. The MONC team would like this to be complemented by an outward-facing web site. Based on our experiences with previous open call projects including BoneJ and Distance, we will prototype a web site template to make it easy for researchers to understand what MONC does, how to use MONC, how MONC works, and the ways in which they can contribute to MONC's future development.
Encouraging the uptake and development of MONC
MONC will enable current and future generations of scientists to simulate atmospheric flows and clouds using modern super-computers. Indeed, MONC is already used by researchers funded by the National Environmental Research Council (NERC) at the Universities of Leeds and Exeter, and there is also strong interest from Reading University.
Our collaboration contributes to encouraging the uptake, and ongoing development, of MONC, within the UK, to replace LEM as the UK's de facto community code for atmospheric sciences, and to allow this community to exploit state-of-the-art super-computer facilities and so to generate more detailed, and accurate, climate forecasts.