PhD: 3D radiation in clouds and wildfire plumes using GPUs and machine learning

This PhD project is part of ORCRIST: Optimized Ray-tracing for Cloud-Radiation Interaction Simulations in 3D using GPUs and Machine Learning, funded by the NWO Open Technology Programme.

Sunlight and heat radiation move through the atmosphere in three dimensions, yet almost every weather and climate model still treats radiation as a purely vertical process. We want to change that. As a PhD candidate in Wageningen, you will help build the first Monte Carlo ray tracer for the atmosphere that is fast enough for forecasting, and use it to find out how 3D radiation shapes clouds, solar energy availability, climate sensitivity and wildfire plumes. You will work in a tandem with a fellow PhD student in computer science at Leiden University, who focuses on GPU optimization, while you lead the physics. Together you will develop a machine-learning denoiser that brings the ray tracer to operational speed. Your research has a clear route to society: throughout the project you will work with a user committee of KNMI, weather forecasting company Whiffle, TNO, software company VORtech and Columbia University.

Your duties and responsibilities include:

• Setting up three benchmark simulations in the GPU-accelerated model MicroHH: solar energy forecasting, tropical climate, and a real wildfire plume.
• Developing a 3D thermal radiation solver within the RTE-RRTMGP-CPP ray tracer, including smart strategies for sampling radiation sources.
• Quantifying how 3D radiation changes clouds, surface irradiance and plume dynamics compared to conventional 1D radiation.
• Co-developing a physics-informed machine-learning denoiser with your Leiden colleague, and evaluating it together with the user committee in operational settings.
• Publishing your results in leading journals and presenting them at international conferences and to users from the field.

You will work here
The research is embedded within the chair of Meteorology (Meteorology and Air Quality Group), led by Prof. Vilà-Guerau de Arellano. You will be supervised by Dr.ir. Chiel van Heerwaarden and Dr.ir. Menno Veerman, both at WUR, and by Dr. Ben van Werkhoven at Leiden University. The team is small, open and collaborative, and has a track record of building widely used open-source simulation codes. You will have access to national and European supercomputers (Snellius, LUMI) and dedicated GPU hardware.