We welcome applications for an 8-week summer placement project on the topic of “Wildfire spread prediction using deep learning and remote sensing data”, with Imperial’s Undergraduate Research Opportunities Programme, on the topic of wildfires.
This UROPs is offered on a remote or hybrid basis. On-campus attendance will be possible, but it will not be essential.
*Applications are especially welcomed by students of Black heritage, as well as from individuals who are members of current and historically underrepresented groups*
Please check the Imperial College London UROP pages for general information and eligibility
A bursary is available, of £378/week to the successful applicant.
“It has been fantastic to learn so much about climate physics and modelling through this placement, and the people at the centre have gone out of their way to make this experience enjoyable and useful for me. I thank them especially for enabling me to participate in the journal clubs and the risk and resilience workshop, both of which I thoroughly enjoyed, not least because they exposed me to areas of research I had never been able to hear about before.” Clara Bayley – previous UROP student with Leverhulme Wildfires in the Dept. of Physics. Read about Clara’s research here.
Project title: Wildfire spread prediction using deep learning and remote sensing data
The motivation for this study is to address the increasing threat of wildfires, which can have devastating impacts on communities, the environment, and public health. With longer fire seasons and more severe fires, there is a pressing need for more effective wildfire management and prediction technologies. Accurately predicting the likelihood and spread of wildfires can enable better land management decisions, disaster preparedness, and emergency response.
This study aims to address the urgent need for novel wildfire warning and prediction technologies, given the significant impact of wildfires on human health, the environment, and properties. The study focuses on predicting wildfire spreading using deep learning models with a newly curated dataset, the “Next Day Wildfire Spread,” which combines historical wildfire data with 11 observational variables overlaid over 2-D regions at 1 km resolution. The resulting dataset has 18,545 fire events, for which two snapshots of the fire spreading pattern, at time t and t + 1 day, are provided. This dataset fills a critical gap in the availability of publicly available wildfire datasets, providing extensive spatial and temporal coverage, and a wide range of variables required for fire prediction. The “Next Day Wildfire Spread” data is available at https://www.kaggle.com/datasets/fantineh/next-day-wildfire-spread.
Different layers of available data used to predict the fire evolution are illustrated in the Figure below (from Huot et al, 2022).
Skills and experience required:
Excellent python programming, experience with Pytorch (preferred) or tensorflow, good knowledge about deep learning, knowledge about constrastive learning is a plus.
Contact details and How to Apply:
- The supervisors will be Dr. Sibo Cheng and Dr. Rossella Arcucci. Day to day supervision involves Mr. Che Liu, current PhD student.
- To apply, please download an application form from here and email the completed form along with your CV to Dr Sibo Cheng (firstname.lastname@example.org) by 30th April 5pm BST.
- This UROP is offered on a remote or hybrid basis. On-campus attendance (South Kensington campus) will be possible, but it will not be essential.
- A bursary is available, of £378/week to the successful applicant.
Please check the Imperial College London UROP website for general information and eligibility
Huot, Fantine, R. Lily Hu, Nita Goyal, Tharun Sankar, Matthias Ihme, and Yi-Fan Chen. “Next day wildfire spread: A machine learning dataset to predict wildfire spreading from remote-sensing data.” IEEE Transactions on Geoscience and Remote Sensing 60 (2022): 1-13. https://towardsdatascience.com/understanding-contrastive-learning-d5b19fd96607