Modern Software Development for Atmospheric Parameterizations and Sub-Grid Scale Models
DescriptionClouds, precipitation, aerosol and turbulence are major sources of uncertainty for global Earth System Models (ESM). The relevant length and time scales are smaller than what can be resolved by ESMs, making parameterizations essential. For some of the parameterized processes our scientific understanding is still limited, while for others the sheer observed variability makes the parameterization design a challenge. The growing available computational resources will not solve those problems any time soon, but offer new possibilities, provided the research software can be easily adapted to the new architectures. The growing public demand for robust and actionable climate forecasts highlights the need for model uncertainty quantification and quality control. I work at CliMA where I'm responsible for the development of new unified sub-grid scale model of turbulence, convection and clouds including the cloud microphysics and aerosol processes. I want to share the lessons I learned on: (i) the benefits of unified sub-grid scale parameterization stack, (ii) documentation driven development, (iii) when to re-implement and when to code something new, (iv) collaborations between software engineers, scientists and students, (v) speeding up development by carefully chosen automated suits of tests, (vi) using machine learning in parameterization design and development.
TimeMonday, June 2616:30 - 17:00 CEST
Climate, Weather and Earth Sciences