Enhancing vital climate data with Bristol’s BluePebble HPC and Machine Learning 

Learn how PhD student Emily Vosper used Bristol’s BluePebble HPC to identify the best machine learning model for enhancing tropical cyclone rainfall data. Her research is helping us better understand the behaviour of these severe storms in the context of climate change.  

In the UK, we are used to heavy rainfall, but not at the intensity of a tropical cyclone.   

As climate change brings about more severe tropical cyclones, enhanced climate data can help us develop more precise predictions of how cyclone rainfall changes will impact local communities. 

Research overview 

Emily Vosper is one of over 1600 Bristol researchers who use the University’s HPC facilities in their research.   

Her PhD research lies at the intersection of Geographical Sciences and Computer Sciences. It examines the use of machine learning models to better understand tropical cyclone rainfall in the context of climate change.  

After extensively testing three different machine learning models on cyclone rainfall data using Bristol’s BluePebble HPC facility, Emily was able to identify which is the best one in terms of computational efficiency and speed.  

Applying the superior Wasserstein GAN model to large weather data sets, she developed a method that can enhance the resolution of cyclone rainfall data from 100 km to just 10 km. These more granular datasets are crucial for assessing the risks associated with extreme weather events and for developing climate adaption strategies. 

How the BluePebble HPC delivered research answers 

Testing these machine learning models involved processing global climate simulations of the past century to the year 2100, extracting the tropical cyclone rainfall and training models on low- and high-resolution pairs. It’s inherently vast and complex simulations.  

The primary challenge of this is the sheer volume of data, which traditional computing resources can’t store nor handle efficiently. BluePebble HPC was able to solve both these challenges.  

Emily used BluePebble for:  

  • Data storage: 4 TB of climate data   
  • Data Processing: Leveraging BluePebble’s powerful GPUs to run machine learning algorithms on large datasets, execute calculations and generate more granular rainfall data.  
  • Machine Learning: Utilising Blue Pebble’s GPUs to evaluate machine learning models and identify the best one in terms of computational efficiency and speed. 

Support from ACRC Staff 

The Advanced Computing Research Centre (ACRC) staff has supported Emily throughout her PhD studies.  

“Taking the Introduction to Python and Introduction to HPC course helped me get started with my research and was very useful. The ACRC team also provided critical assistance in troubleshooting code and helped me optimise my code so I could use the HPC resources effectively.  

Their support was also essential in maintaining workflow efficiency and meeting tight research deadlines”, Emily Vosper, PhD candidate. 

Future research and applications  

The machine learning method Emily developed in her PhD research has potential applications beyond tropical cyclone rainfall.  

Adapted to other climate data sets, such as temperature and other storm types, the method could give us a more granular prediction of how climate change will increase flood risk and heat-induced mortality, so we can better prepare for future climate scenarios.  

Advance your research with HPC  

Bristol’s HPC environments and research data storage facilities are open to all staff and students. 

Get started using the University’s digital research infrastructure:  

 

View all IT for research services and support 

 

 

Edit this page